Cantor Fitzgerald, Silverstein Close $176M Construction Loan – Multi-Housing News
Affiliates of Cantor Fitzgerald and Silverstein Properties have closed on a $176.2 million construction loan for the development of Astra Tower, a 40-story, 377-unit residential building in Salt Lake City that is the latest investment by Cantor Silverstein Opportunity Zone Trust Inc.
Cantor and Silverstein, along with their partner Kensington Investment Co. Inc., secured the construction loan from Mack Real Estate Credit Strategies for the development that will be the tallest building in Utah upon completion. Excavation and foundation work began in the spring with delivery of the high-rise expected in late 2024.
Kensington, a Boston-based developer, broke ground on the project in January. Ed Lewis, CEO of Kensington Investment Co., said in prepared remarks the company believes Astra Tower will elevate Utah multifamily and create a new gold standard for apartment living throughout the West.
Designed by HKS, Astra Tower will be located at 200 S. State St. and stand 450 tall. It will offer studio, one- and two-bedroom apartments, with two floors of penthouse units. Amenities will be spread among three floors. The eighth floor will feature a terrace, a gourmet kitchen, a lounge with televisions and games, a fireplace, an outdoor deck with seating and grills, an urban park and fitness center and fitness studio.
The 22nd floor will have a pool terrace with a swimming pool, jacuzzi, spa with saunas, plunge pool and treatment rooms. The 40th floor will be a residents-only rooftop terrace with a big-screen TV, a grill and group socialization areas.
The development team is seeking LEED Gold certification and is planning to integrate WELL Building concepts into the design to enhance the health and well-being of residents.
Jacobsen Construction Co. is the general contractor.
Cantor and Silverstein launched an Opportunity Zone business in early 2019 to invest in, develop, redevelop and manage a diversified portfolio of institutional quality commercial real estate assets with an emphasis on multifamily properties in Qualified Opportunity Zones in the United States.
The current OZ development portfolio includes nine projects across seven states comprised of 2,788 residential units and a 250,000-square-foot life science facility in Philadelphia totaling $2 billion in estimated project costs. In addition to Astra Tower, other CSOZ Trust multifamily projects include:
The latest multifamily news, delivered every morning.
Like what you’re reading? Subscribe for free. Subscribe
© 2022 – Multi-Housing News Online
Multi-scale-average-filter-assisted level set segmentation model with local region restoration achievements | Scientific Reports – Nature.com
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 15949 (2022)
Metrics details
Segmentation of noisy images having light in the background it is a challenging task for the existing segmentation approaches and methods. In this paper, we suggest a novel variational method for joint restoration and segmentation of noisy images which are having intensity and inhomogeneity in the existence of high contrast light in the background. The proposed model combines statistical local region information of circular regions centered at each pixel with a multi-phase segmentation technique enabling inhomogeneous image restoration. The proposed model is written in the fuzzy set framework and resolved through alternating direction minimization approach of multipliers. Through experiments, we have tested the performance of the suggested approach on diverse types of synthetic and real images in the existence of intensity and in-homogeneity; and evaluate the precision, as well as, the robustness of the suggested model. Furthermore, the outcomes are, then, compared with other state-of-the-art models including two-phase and multi-phase approaches and show that our method has superiority for images in the existence of noise and inhomogeneity. Our empirical evaluation and experiments, using real images, evaluate and assess the efficiency of the suggested model against several other closest rivals. We observed that the suggested model can precisely segment all the images having brightness, diffuse edges, high contrast light in the background, and inhomogeneity.
In the domain of image processing, the two terms image segmentation and image restoration are closely related tasks with various applications in engineering, technological fields, pathology, astronomy, advanced driver assistance systems, etc. Utilizing aspects of image restoration to efficiently segment objects with, relatively, a high degree of noise, blur, missing pixels, or inhomogeneity is a commonly encountered task. Restoration and segmentation task can be simultaneously combined into a scheme or jointly represented into a minimization functional. Due to the specific properties of the images and segmentation requirements, the combined joint schemes result to be more useful and accurate.
The restoration task was the first task encountered after photography development. The first restoration approach was towards noise removal followed by deconvolution, inpainting etc. One of the most successful models for additive noise removal was total variational (TV) based model introduced by Rudin Osher Fatemi (ROF)1 in 1992. The task of image denoising restoration is the evaluation of a desired original image (u({textbf {x}}):Omega longrightarrow mathbb {R}) given (u_0({textbf {x}})=u({textbf {x}})+eta ({textbf {x}})) with (u_0({textbf {x}}):Omega longrightarrow mathbb {R}) corrupted image with additive noise (eta ({textbf {x}})). To reconstruct (u({textbf {x}})) from the observed degraded image (u_0({textbf {x}})), the ROF method utilizes the TV norm as a data regularization term. The energy functional of the ROF model1 is illustrated mathematically in the following Eq. (1):
where (lambda ) is a regularization parameter, (Omega ) is the image domain and (int _{Omega }|nabla u({textbf {x}})|d{textbf {x}}) characterizes the total variation of (u({textbf {x}})). The minimization of the Eq. (1) leads to a PDEs grounded methods and approach which has been additionally drawn-out and revised to the separation of multiplicative noise through implementing the logarithm transformation or other restoration techniques. In fact, one of the key and foremost advantages of the ROF model1 is that it conserve and store the edge information while smoothing the leftover regions2. This quality is of help in case the image will further be segmented. In order to remove the staircase effect or better performance different methods have been introduced3,4,5,6,7,8. Most of these techniques use high order derivative terms to obtain more details and reduce the staircase effect. The data fidelity term (int _{Omega }(u_0({textbf {x}})-u({textbf {x}}))^2d{textbf {x}}) used for ROF model shows effectiveness in case of Gaussian additive noise. Note that, for removing other types of noises, such as Poisson noise, impulse noise, etc., the fidelity term must change accordingly9,10. The denoising problem can be extended to image restoration by introducing a blurry linear operator (varphi ) such that (u_0({textbf {x}})=varphi ({textbf {x}}) u({textbf {x}})+nu ({textbf {x}})). As the restoration problem has two unknown, (varphi ) and u, dealing with it is harder. For image restoration problem, local spatial convolution filters of both one-dimensional and two-dimensional signals filters in combination with denoising techniques has been suggested and designed based on the quantitative properties of the image.
On the other hand, image segmentation is, in fact, one of the basic and crucial tasks in the arena of image processing. This should be noted that image segmentation mainly focuses to split the presented image into various meaningful regions and, subsequently, obtain some useful information11,12,13, such as the distinction between foreground and background or object/feature separation. In general, image segmentation is largely implemented as a pre-processing and/or post-processing technique combined or coupled with other techniques such as image restoration, image/pattern recognition, etc. In this paper the word “inhomogeneous” refers to such images in which objects pixels values changes slightly from one side to another means having no fix value for an entire object or objects having diffuse edges which make segmentation problem difficult. The “Global information” means the average of all pixels values of an image/object. While “statistical local region information” refer when we take K (times ) K window size matrix from an image and dealt with it by using statistical terms like mean, variance etc. In considering local information we take small region information which are usually taken in the form of circle using simple circle equations that is referred as circular regions. Moreover, we implement and make use of the level set approach which is assumed as a powerful numerical technique for image segmentation and analysis. In order to tackle with all the aforementioned problems we introduce a new model which consists of:
coupling into an energy functional the desired restore image and the filtered image into a single energy functional for segmentation purposes using a fuzzy membership function;
extending the Cai model14 to a new variation segmentation approach that has the capability to restore inhomogenous images by adapting image global information to statistical local region information of circular regions centered at each pixel with a multi-phase segmentation technique;
introducing a variational image segmentation approach and image restoration approach which are capable to handle the segmentation of images with high contrast light in the background;
we propose a new model which takes the advantages from both restoration, as well as, image filtering for segmentation purposes using the fuzzy membership function in difference with Cai model14, which utilizes the restored image obtained from ROF formulation; and
through wide empirical analysis on different classical datasets of images, we ascertained that the suggested model is precise and effective, particularly, in images that have brightness, diffuse edges, high contrast light in the background, and inhomogeneity.
The remainder discussion if this paper is organized as deliberated in next sentences. In ”Related work” section, we offer a summary and overview of state-of-the-art image segmentation techniques. In “Machine learning for image segmentation” section, we deliberate various machine learning techniques and their role in the image segmentation. In “The proposed multi-scale-average-filter-assisted local region restoration segmentation ((M_{SAF}L_{RR}S)) model” section, we detail the suggested revised and new segmentation model and alter it to vector-valued images. In “Experimental results” section, we assess and evaluate the performance of the suggested approach with other state-of-the-art segmentation approaches using different real-world and synthetic images datasets. In the assessment, we consider various images having brightness, diffuse edges, high contrast light in the background, and inhomogeneity. Conclusions are drawn in “ Conclusions and future work” section along with several guidelines for future investigation and research.
One of the most ambitious and stimulating tasks in image segmentation is intensity inhomogeneity. Problems, for instance, artificial illumination and non-uniform daylight can cause imperfection of acquisition which leads to image inhomogeneity. Intensity inhomogeneity highly affects the image segmentation precision due to the overlap of background and foreground. In the last decade, many promising algorithms and methods were introduced to tackle this problem15,16,17,18,19,20. However, all those methods have limitations and are unable to tackle severe intensity inhomogeneity and work for images with specific properties21,22. For a better understanding of their limitation, we will shortly revise and comment on some state-of-the-art approaches and techniques. In summation to the intensity-based approach, we will discuss the widely used deep learning-based methods and well-known techniques for tackling the low-level computer vision problems, for instance de-noising and artifact removal.
The main image segmentation techniques can be classified as: (i) edge-based segmentation approaches; and (ii) region-based segmentation techniques. The edge-based models13,23,24 incorporate edge detector functions which channelize the movement of active contour in the directions of the boundaries. Such functions rely on the gradient of the image data. The region-based models11,16,17 utilize region information, such as variance, mean etc., to move the contour in the directions of the object’s boundary. The edge-based models utilize the image local information to be unable good performance in noisy images and ignore the objects having diluted boundaries. On the other hand, the region-based methods and models are unable to tackle the intensity inhomogeneity. This is due to the fact that the intensity inhomogeneity is the local property of images rather than a global one. One of the benefits of region-based techniques is that these approaches are potentially less sensitive to the noise and outliers due to which their segmentation results are better in noisy images. In fact, the majority of the region grounded models are approximations to the milestone Mumford–Shah (MS) energy functional12. Among all of them, active contour without edges suggested by the Chan and Vese (CV)11,12 gained much popularity in the literature due to its simple implementation. In terms of the CV model, the energy functional is given and illustrated in Eq. (2):
where (u_0({textbf {x}})) is the given image, (Gamma ) denotes smooth and segmented curve, (mu ), (lambda _1) and (lambda _2) are positive parameters (to be tuned accordingly), (c_1) and (c_2) are the mean intensities of (u_0({textbf {x}})) inside and outside of the (Gamma ), correspondingly. Although, the CV model is commonly used and it has promising results for additive Gaussian noise, however, its limitation can be easily observed in cases the image suffers from intensity inhomogeneity25,26. This drawback is due to the utilization of the global information of images and ignoring local features information27. To enhance the CV model for inhomogeneity image segmentation the Local Binary Fitting (LBF) model16 was introduced. The LBF model hires a kernel function to locate the local intensity information of images and embeds this information and statistics into a region-based active contour model and level set formulation28,29. In terms of the LBF model, the energy functional is given as illustrated in Eq. (3):
whereas the variables (lambda _1), (lambda _2) are constants, (K_{sigma }) represents the Gaussian kernel with standard deviation ((sigma )). Furthermore, the variables (g_1) and (g_2) characterizes the two smooth functions that, in fact, approximate the local details and statistics of the image inside and/or outside of the (Gamma ), correspondingly. Although, the LBF model can cope with the intensity inhomogeneous; nevertheless, this model is very sensitive to the initial contours. Moreover, changes on initial contour can potentially lead the LBF model to produce undesirable segmentation results. Therefore, to further improve the segmentation of intensity inhomogeneity images and for bias field correction, Li et al.30 suggested a new region based variational model31,32. The authors in Ref.30 defined an objective function for K-means clustering, which is weighted, in a locality close to every point, with the centers of the clusters and having a multiplicative component that, in fact, computes and estimates the bias within the locality. Subsequently, then the proposed function is amalgamated over the whole environment and embedded into a level set formulation. Even though, the method suggested by Li et al.30, overcomes the existing ones, still the method can not deal with high image inhomogeneity, as similar to the other methods, the method is grounded on the laying claim that every intensity inhomogeneous image is, in fact, homogeneous within a small region. Another problem with those methods is that there is no prediction on the scale of the homogeneous region. Dealing with serious and hard intensity inhomogeneity and tuning the scale, in particular, for inhomogenous regions may potentially cause undesired results. Taking into account these problems, Wang et al.20 suggested a multi-scale local (MSL), and region oriented, system and model for segmentation of intensity inhomogeneous images. With the assumption that the desired neat image (u({textbf {x}})) is vitiated and damaged by the additive noise (eta ({textbf {x}})) and the intensity inhomogeneity (varphi ({textbf {x}})), then the obtained image (u_0({textbf {x}})) is described as given by Eq. (4):
A generally accepted assumption is that the intensity inhomogeneity is a tardily changing and varying component over the entire image and it is constant within a small local region. The target is to acquire the corresponding clean image (u({textbf {x}})) which is impressed and disturbed by both the noise and intensity inhomogeneity. To achieve this, the MSL model defines a local region in a circular shape for capturing local information and statistics and then some kind of mathematical and statistical assessment is done on those local circular arenas for each and every pixel utilizing multi-scale low-pass filtering. Assuming (hat{u}({textbf {x}})=varphi ({textbf {x}}) u({textbf {x}})), we have the following relationship:
Applying Eq. (1) it is easy to recover (hat{u}({textbf {x}})) and then consider it as a given image which is only suffered from intensity inhomogeneity and free from noise. Thus, the problem reduces into finding (u({textbf {x}})) from (hat{u}({textbf {x}})=varphi ({textbf {x}}) u({textbf {x}})) where (varphi ({textbf {x}})) is the intensity inhomogeneity. After applying the logarithmic transformation, we obtained:
As both the inhomogeneity layer (varphi ({textbf {x}})) and the clean image (u({textbf {x}})) are unknown directly finding the clean image (u({textbf {x}})) from Eq. (6) it is impossible. To defeat this difficulty. Wang et al.20 suggested a multi-scale average filter. The local circular regions are defined in order to make the model more adaptable in capturing intensity information in the local region of a given pixel. To examine and investigate the information of the local circular region at each center pixel (mathbf {x}) of the, particular, given image (hat{u}) the multi-scale average filter is designed in the following form as illustrated in Eq. (7):
where, the subscript i is the radius of the local circular region and this can also be characterized as a scale parameter. Furthermore, the variable n symbolizes the total amount of pixels within that particular local circular region (F_{mathbf {x},i}) with center (mathbf {x}); and is subsequently defined by the following Eq. (8):
Furthermore, (M_k(mathbf {x})) is taken to be the mean of the multi-scale average filter and it is characterized as illustrated using Eq. (9):
where k represents the entire amount of the scales and this needs to be tuned properly according to the images. It may be noted that in case the value of the variable k is little then very elite circular regions will be investigated for every center pixel which may lead to an unfavorable result. Similarly, on the other hand if in case the value of the variable k is taken very ample then it will potentially increase the computational cost due to the fact that too many local circular regions will consider for every center pixel value. By replacing (varphi (mathbf {x})) in Eq. (6) by (M_k(mathbf {x})), we get the following relationship:
In fact, this should be noted that (bar{u}) is an approximation to the clean image or intensity inhomogeneity free image u, whereas (M_N) is a constant, in fact normalized, to conserve the mean intensity of (bar{u}). Furthermore, Eq. (10) can be represented in an equivalent form to decrease the computational cost as:
In fact, Eq. (11) represents an approximation of inhomogeneity-free image and shows that (bar{u}) can be obtained by dividing (hat{u}(mathbf {x})M_N) by multi-scale intensity information (M_k(mathbf {x})). The filter can be named as dual filter formulation, as the image has been filtered twice and then divide it by its average. For a better understanding of the dual filter formulation, we show experimental outcomes, in particular, for a gray-scale synthetic inhomogeneous image and, subsequently, a color image of the plane with relatively high brightness within the background, as shown in Fig. 1. The dual filter is implemented on these two test images with k value 10, 20 and 30, respectively, as shown in Figs. 2 and 3.
Two test images: the first one is gray-scale which is suffered from intensity inhomogeneity while the second one is color image having brightness in the background. These inhomogeneous images are taken from Cai et al.14 and are publicly available at the kaggle website33.
From Fig. 2, first row third column, it is clear that the intensity inhomogeneity is almost covered but at the same time the edges are also diffuse and an extra region around edges become darker which may cause an unsatisfactory result in segmentation. When the value of k increases from (k = 10) to (k = 20) and then (k=30) we notice that the inhomogeneity is slightly removed for higher values of k as clear from Fig. 2a,c, but the edges are not affected and an extra region around the edges is not damaged. Figure 3 demonstrates the results of dual-filter formulation on a real-world color image of a plane, which has high brightness in the background due to sunlight which may cause difficulty in segmentation. On the other hand using the filtered image instead of the original one makes the segmentation task easier and more efficient as filter images are clear from the original one. However, it may be noted that the scale parameter k plays a vital role in the dual filter formulation, shown in the last column of Fig. 3. As the value of k increases, we notice more content in the resulting image but at the same time, it increases the computational cost. Through experiments, we noticed that the k value can vary from 5 to 35 and default (k=30) is more appropriate.
A demonstration of the dual filter formulation on gray image suffered from intensity inhomogeneity with different values of k. The first column (a) shows the inhomogeneity covered by first filter, the second column (b) shows the inhomogeneity covered by the second filter, and the last column (c) shows the inhomogeneity-free images. Furthermore, the first row deliberates the outcomes for (k=10), the second row deliberates the outcomes for (k=20), and the last row demonstrates the results for (k=30).
A demonstration of the dual filter formulation on color image of plane having high brightness in background, with different values of k. The first column shows the inhomogeneity covered by first filter, the second column shows the inhomogeneity covered by the second filter, and the last column demonstrates the inhomogeneity-free images. The first row deliberates the outcomes for (k=10), the second row deliberates the outcomes for (k=20), and the last row demonstrates the obtained results for (k=30). These inhomogeneous images datasets are collected from the Goldstein et al.6, Cai et al.14 and are publicly available at the kaggle repository33.
Alternatively, deceptively regularized kernel-oriented techniques have been used to enable local information into the segmentation fitting term. For instance, Elazab et al.34 proposed a deceptively regularized and fuzzy kernel-oriented C-means clustering (ARKFCM) system and framework. The suggested system has applications in terms of segmentation capabilities for brain MR images and inhomogeneous datasets with the energy function as illustrated mathematically in Eq. (12):
with (x_i), (i=1:N), image gray scale in k dimensional space, (v_j), (j=1:c), cluster center, (u_{i,j}), the membership value for every pixel i and (j{text{-th}}) cluster, and K the Gaussian radial basis function. Note that in this framework three different algorithms have been suggested that consists of: (i) the local average gray-scaling being substituted by the gray-scale of the mean filter (ARKFCM(_a)), (ii) median filter (ARKFCM(_m)), and (iii) devised weighted images (ARKFCM(_w)), correspondingly. In fact, all these algorithms utilize the heterogeneity of the gray-scales in the pixel locality and, subsequently, put to work this assessment criterion for local contextual information. This should be noted that this is achieved by replacing the standard Euclidean distance with the Gaussian radial basis kernel functions. The ARKFCM framework is independent of parameters which is one of the main advantage of this method and also has promising results for images in presence of noise. Although, one can observe limitation of this technique with images having intensity inhomogeneity, as will be later shown in Fig. 10, usually occurring in MR images.
Recently, Cai et al.14 suggested a variational framework for image segmentation while taking advantage of the image restoration techniques. In this work a link among the image segmentation and the image restoration approaches has been shown, whereas Cai et al.35 proves arguments on the fact that the solution of CV model11 can be achieved through thresholding the minimizer of the ROF model1. Finally, the energy functional of the Cai et al. model14 is based on two different data fitting terms, i.e., (i) one for image restoration, and (ii) the other one for image segmentation. The relationship is illustrated mathematically as given in the following Eq. (13):
where (sum _{i=1}^K v_i(mathbf{{x}})=1,v_i(mathbf{{x}})in {0, 1}) is a fuzzy membership function, and (mathcal {O}) is a blurring operator i.e., if blur is observed in the image and the aforementioned is a recognition operator for a noisy ascertained image, as well. The blurring operator (mathcal {O}) can be computed through using various image de-blurring methods and techniques as suggested in Refs.36,37,38. The Cai et al. model14 can efficiently segment images that are damaged and corrupted with the high noise, blur affect, and/or missing pixels; however, its limitation can still be observed in intensity inhomogeneous images, which represents the main problems in the Cai et al. model14. This drawback of the Cai et al. method14 is due to the fact that the suggested approach and method uses only global information of images and ignores the local one. This issue can be solved through implementing certain machine learning approaches into the image segmentation methods. In next section, we discuss how machine learning based methods can be integrated into existing approaches and use them in the field of image segmentation.
Deep learning-based method is a newly emerged techniques for image segmentation purposes which has so many types and among them the famous one are: (i) convolutional models with graphical models; (ii) fully convolutional networks; (iii) encoder–decoder based models; and (iv) multi-scale and pyramid network based models39. Out of these types, the convolutional neural networks (CNNs) have gained much popularity and extraordinary success in this task of image segmentation. All of these techniques rely on the idea of machine learning approaches and has shown so many promising and excellent results. There are also some method which utilized both active contour and CNN idea to overcome the task of image segmentation, like deep active contour network (DACN) method introduce by Zhang et al.40. However, the problem with the CNN approach is that it performs quite badly at identifying precise object boundaries. The major cause is in fact the information loss in the successive down sampling layers41,42. On the other hand, the active contour models generate relatively more accurate and useful localization of boundaries by fitting an arch for the object shape in the image using a series of approaches, for instance, (i) the edge based, and (ii) the region based techniques.
In addition to the intensity-based approach, we will discuss the deep learning-based method for tackling low-level computer vision problems, such as de-noising and artifact removal43. In Ref.44, the authors demonstrate that CNN-based de-noising algorithms, in fact, try to ascertain a mapping function through the maximization of a loss function over a training dataset of the degraded-clean picture pairings. The temporal complexity of the acquisition procedure is quite eminent, despite the fact that this approach is successful and has a short running period. The use of a hierarchical network has improved the learning of high-level features, and thanks to the development of the CNN-based de-noising algorithms. For image segmentation challenges, the Mask RCNN is a deep learning model which is largely used in the existing literature. In fact, this can use the bounding box, classes, and binary image mask to separate various pictures in an image or video. Furthermore, the Faster RCNN was used to create Mask RCNN. For each candidate, F-RCNN produces two outputs: (i) a class label; and (ii) a bounding box. In this paper, we undertake the key and fundamental issue of the intensity inhomogeneity and back light struck images which are still a challenging task for CNN methods as they rely on image average pixels’ values while intensity inhomogeneity is a local property of an image rather than global/average one. Therefore, we rely on active contour along with fuzzy membership function in this article.
Similarly, the long and short term model (LSTM) are also widely used for image segmentation43. However, traditional LSTM models are not good because they cannot capture the spatial information of images. Similarly, fully connected weights may significantly increase models computational costs. Therefore, instead of traditional LSTM models the convolutional LSTM methods have been largely used to perform instance-level segmentation. These models can choose every instance of the object in dissimilar timestamps of the sequential result and output. Therefore, to further improve the model performance, attention models are guaranteed as they are supposed to have higher control over the operation of localising particular instances than the traditional convolutional LSTMs, which might choose distinct instances of objects at different timestamps. In Ref.45, a deep learning-based denoising technique is demonstrated that incorporates the CNN model with residual connection and attention mechanism. After the Attention-Residual mechanism has calculated the amount of noise in the image, it may be further removed using a simple additive procedure, resulting in the denoised image. A summary of various deep learning based models, including RNN-based methods, can be found in Refs.39,43,46.
Inspired by the well-known Cai et al. model14, we suggest a new and novel extension of it which integrates both the global and the local region information into the segmentation of images suffering from intensity inhomogeneity and noise. The proposed work profits from both restoration, as well as, image filtering for segmentation purposes using fuzzy membership function. In difference with Cai model14, which utilizes the restored image obtained from ROF formulation, we use multi-scale-average-filter to enable to deal with inhomogeneity toward a accurate image segmentation in the level set formulation. In our proposed method we utilized two types of images namely, (hat{u}) and (bar{u}) both these images are actually approximation to the given noisy and intensity inhomogeneity image (u_0). Here (hat{u}) represent image free from noise and (bar{u}) represents intensity inhomogeneity free image obtained through dual filter formulation. In difference with the Cai et al.14 model, which utilizes the restored image obtained from ROF formulation, we use multi-scale-average-filter to enable to deal with inhomogeneity toward accurate image segmentation. By doing so our model is not only capable to tackle noise but also tackle images having diffuse edges, light in background and sever intensity inhomogeneity. In contrast, Cai model is not able to tackle such kind of images as shown and discuss in details in experimental section. The energy functional of the (M_{SAF}L_{RR}S) model is illustrated in Eq. (14):
where (sum _{i=1}^K v_i(mathbf{{x}})=1,v_i(mathbf{{x}})in {0, 1}), (forall xin Omega ,) (hat{u}in L^2(Omega )) and (mathcal {O}) is a linear operator. In this paper, we consider (mathcal {O}) as a Gaussian kernel as we have been dealing with Gaussian noise. The first term in Eq. (14) is a image restoration data fitting term. This term removes noise and also controls the closeness of the function (hat{u}) to the given image (u_0). The second and the third terms consist on two fitting terms which utilize both the recover image (hat{u}) and the dual-filter image (bar{u}) (obtain from Eq. (10)). The third term aims to segment the given image into K different intensity levels whereas the second term supports the third term with a fitting to inhomogeneity free clean image. The last term is a TV smoothing term.
This should be noted that the prolongation and revision of the suggested system and model to vector-valued images format is straight forward. Let (mathbf {u_0}=(u_{01},cdots ,u_{0p})), (hat{mathbf {u}}=(hat{u_{1}},ldots ,hat{u_{p}})), (bar{mathbf {u}}=(bar{u_{1}},ldots ,bar{u_{p}})), (mathbf {k_i}=(k_{i,1},ldots ,k_{i,p})) and (mathbf {c_i}=(c_{i,1},ldots ,c_{i,p})), then model (14) can be extended for segmenting vector-valued images as:
Before minimization of the functional (14) we first relax (v_i) as; (sum _{i=1}^K v_i(mathbf{{x}})=1,v_i(mathbf{{x}}) ge 0,) for all (mathbf {x} in ) (Omega ). Keeping (k_i), (v_i), and (c_i) as constant and deriving with respect to (hat{u}) in Eq. (14) we have:
Similarly, keeping (v_i), (c_i) and (hat{u}) as constant, and minimize (14) with respect to (k_i) one have:
Same procedure can be adopted to find the values for (c_i) by minimizing (14) with respect to (c_i), for fix (v_i), (k_i) and (hat{u}):
As discussed in Ref.14 that it is possible, and we can prove mathematically, as well as, theoretically, that if (sum _{j=1}^pint _{Omega }(u_{0j}-mathcal {O}_jhat{u}_j)^2dmathbf {x}) is convex and continuous, then for fixed (k_i) and (v_i) there exists only one u which minimize the energy consumption using Eq. (14). In order to find (v_i) with fixed (hat{u}) many methods can be adopted, such as Alternating Direction Method of Multipliers25,47,48, or applying the primal-dual algorithm8,49, or the max-flow approach50. For more inside information about finding and estimating the value of (v_i), interesting readers are referred to Ref.14. The steps of the suggested approach is given as follows in Algorithm (1):
In this section, we demonstrate and perform numerous numerical experiments to measure and assess the performance of the suggested model. We first show the segmentation accuracy of the suggested model across different synthetic and real dataset images suffering from intensity inhomogeneity in the existence of the noise. We show comparison results of (M_{SAF}L_{RR}S) model with state-of-art models, for instance, LBF, ARKFCM and Cai models, and show outperforming of the proposed model for real-world images which show complexity due to intensity inhomogeneity, noise and brightness in background. Through numerical experiments, we validate and confirm that the suggested approach is relatively faster and much precise in segmenting images having inhomogeneity and noise. Note that, all the simulations were taken out in the Matlab (R2009b) software on a system with Intel i3 2.2 GHz CPU, 3G RAM, and running the Windows 8 operating system. In all the experiments the values of (mu ) and (lambda _1) were set constant of value 1 except in experiment (Fig. 10), in which value of (mu ) is taken 50.
The k-scale parameter has been set according to the image noise level and intensity inhomogeneity. The maximum number of iterations in each experiment for (M_{SAF}L_{RR}S) model has been set 100 and the size of images in the range (250 times 250) for all the experiments except experiment 9 where the size of image is (150 times 150) due to the fact that LBF model produces extra computational cost. The datasets and images used during the experimental study are publicly available in the kaggle repository, and can be accessed online at (https://www.kaggle.com/datasets/mnavaidd/image-segmentation-dataset). This should be noted that we measure the correctness of the suggested model through the factor of Jaccard similarity coefficient51 and the Sørensen–Dice similarity index, as described in “Sørensen–Dice similarity ” section. In other word, this means that we can assess and measure the similarities among the ground truth X and the obtained image Y using the Jaccard index. This index is numerically formulated by the following Eq. (19):
where J denotes the Jaccard index. The Sørensen–Dice similarity index and its method of computation is described in “Sørensen–Dice similarity” section.
In order to validate and quantify the performance of the suggested model, we start with a synthetic blood vessel image that is affected from a slight intensity inhomogeneity and has a presence of noise, as shown in Fig. 4a. The Gaussian noise to the blood vessel image varying as: case (1) zero mean and variance 0.01, and case (2) mean 0.3 and variance 0.02, as given in Fig. 4b,c, respectively. From Fig. 4c–e, we can clearly understand that the suggested approach can perfectly segment those given images. Furthermore, this also is clear and evident from various Fig. 4e,f that our anticipated approach has the potentiality and quality to successfully segment the noisy images, very well, as compared to other state-of-the-art models.
This figure demonstrates the performance of the suggested approach (M_{SAF}L_{RR}S) on a slight intensity inhomogeneous image in which Gaussian noise is added and increased gradually from left towards right and top towards bottom. The first row represents all those given images that suffered from intensity inhomogeneity and noise while the second row represents the corresponding segmented outcomes of the proposed model. The images were taken from the kaggle repository33 and the parameters which were used are (lambda _2=20) and (k=32) (scale parameter).
Figure 5 represent experiments of the suggested technique on four different images having intensity inhomogeneity, brightness in background and texture. For instance, the most first column of Fig. 5 displays the given images, second column shows filter images obtained from dual filter formulation while the last column represents the segmented outcomes that were attained with the use of the suggested method. The images taken into account in the first and the second row of this figure are blood vessel and car number plate images which suffer from intensity inhomogeneity and brightness, the third row displays the image of a jet having average intensity background and low and high intensity objects in the foreground, whereas the last row shows a texture image having three different types of texture regions. In fact, this can be easily understood from the very last column of the Fig. 5 that the proposed approach (M_{SAF}L_{RR}S) has accurately segmented those images.
The above figure assesses the performance and effectiveness of our suggested technique on four different images6,14. The first column (a) shows observed images, the second column, (b) shows filter images, while the last column, (c) shows the corresponding segmented results. The parameters used for the first row are (lambda _2=10) and (k=15), for the second row (lambda _2=30) and (k=10), for the third row (lambda _2=100) and (k=32), and for the last row (lambda _2=1), (k=30). These inhomogeneous images are taken from Goldstein et al.6, Cai et al.14 and are publicly available at the kaggle website33.
In order to evaluate and further assess the restoration and segmentation performance of the suggested approach, we have employed a public fingerprint image dataset for the purpose. The data set is downloaded from Image Processing Place, Bologna dataset (http://www.imageprocessingplace.com/root_files_V3/image_databases.htm), which consists of eighty images of finger prints with unilluminated foreground and diffuse edges. Due to space limitation, we show in Fig. 6 the results of 18 fingerprint images out of 80. In all the experiments of this dataset, the parameters are kept fixed: (mu =10), (lambda _1=5), (lambda _2=300) and (k=5)(scale parameter) and the maximum number of iterations 100. In the first, third and fifth rows of Fig. 6 we represent the original images whereas the second, fourth and sixth rows represent the corresponding segmented outcomes and findings of the (M_{SAF}L_{RR}S) model, respectively.
This figure demonstrates the results of our proposed model on a data set of 18 finger prints images. The images were taken from Ref.33. The first, third, and fifth rows characterizes the original images while the second, forth and sixth rows characterizes the proportionate segmented results, respectively. In all these experiments the values of various parameters were kept constant which are (lambda _2=300) and (k=5) (scale parameter).
We will compare the suggested approach with other state-of-art models including, LBF16, Cai14, and ARKFCM34, and illustrates the experimental results. For a better apprehension of the advantage of the anticipated technique, we start its comparison with the stimulus model of the Cai et al.14. Figures 7 and 8 show some synthetic and real image data (first column) processed with Cai et al. method14 and the proposed method. The results obtained from the Cai et al. method14 are given away in the second column whereas the outcomes obtained with our model in the last column. We can intelligibly see that the suggested approach, similar to the Cai et al. method14, not only can easily deal with such images but furthermore, the method shows a perfect accuracy of the segmented images by including all the object’s details. The images considered in Figs. 7 and 8 suffer from intensity inhomogeneity and having unilluminated objects boundaries. The first row of Fig. 7 is a synthetic image with a bright portion suffered from intensity inhomogeneity. We can clearly understand that the suggested and anticipated approach has successfully segmented the object of this image whereas we see a partially successful result of the Cai et al. model14. The second row of this figure shows the results of synthetic images suffered from slight and severe intensity inhomogeneity, respectively. Form the second row second and third column of Fig. 7 it is clearly seen that both the Cai et al. model14 and the proposed model perform very well in segmenting the overall object of the iron bar but on the other hand, it may be noted that the Cai et al. model14 fails to segment the interior circle in the object while (M_{SAF}L_{RR}S) successfully segments them. The third row of Fig. 7 consists of a multilevel intensity image with grievous intensity inhomogeneity. Nevertheless, in order to segment such an image we use the supposition that the given image has two different levels of intensity of the objects to be segmented. To proceed with the segmentation of such images a multi-phase segmentation can be used, specifically two phase image i.e., the value of K (number of phases) is two for both the (M_{SAF}L_{RR}S) and the Cai et al. model14. We clearly see from the third row in Fig. 7 that the Cai et al. model14 (second column) extracts only two objects while (M_{SAF}L_{RR}S) (third column ) extracts all the three objects. The last image of Fig. 7 is a plant image having diffuse branches. The last row (second and third column)in Fig. 7 shows that the Cai et al. model14 segments only the leaves and is unable to segment its branches while on the other hand (M_{SAF}L_{RR}S) method can capture both of them.
A comparison of the Cai et al.14 and the suggested approach on four dissimilar images that are suffered with slender and grievous intensity inhomogeneity33. The first column represents given images, however the second and third columns display the segmented results of the Cai et al.14 and proposed models, respectively. Moreover, the parameters used for (M_{SAF}L_{RR}S) model are: (c) (lambda _2=10), (k=30), (f) (lambda _2=100), (k=30), (i) (lambda _2=50), (k=35) and for (l) (lambda _2=200), (k=5).
Figure 8 demonstrates the results of the Cai at al.14 and (M_{SAF}L_{RR}S) model on four real-world test images, in which the first image to be tested is a plane image with sun light, the second image is a plane image with diffuse edges, while the third and the last images are helicopter and crescent moon images with unclear edges. In the first experiment of Fig. 8 (shown in the first row) both Cai14 and (M_{SAF}L_{RR}S) models perform very well. In the second row of this figure, we can see that Cai model14 fails to segment the image of the plane with diffuse edges whereas the proposed model has good segmentation outcomes (as given away in the second row last column of Fig. 8). In addition, the last column of the third and the forth row of Fig. 8 clearly demonstrate the robustness of our proposed model (M_{SAF}L_{RR}S) over the Cai et al. model14, i.e., a total fail of Cai model14 for such images. In fact, the proposed model (M_{SAF}L_{RR}S) segments very efficiently the edges as seen in Fig. 8.
In Fig. 9, the first, second, third and fourth column represent the original image, the obtained results with LBF model16, Cai et al. model14, and outcomes of the anticipated technique, respectively. The first row of this figure comparisons the findings for a synthetic image with slight intensity inhomogeneity. In the second and third row Gaussian noise is added with mean zero and variance 0.001 (second row) and 0.005 third row, respectively. From the last column in Fig. 9, we can notice that (M_{SAF}L_{RR}S) performs very well in all the three test images as compared to the LBF16and the Cai et al. models14. The imperfect results of the LBF16 and the Cai et al. model14 can be observed clearly in second and third column, correspondingly.
This figure illustrates the results of the Cai et al.14 and the suggested approach on four dissimilar color images having brightness, diffuse edges and inhomogeneity6. The first column represents given images, while the second and third columns display the segmented results of the Cai et al.14 and the proposed models, respectively14. Note that the parameters used for (M_{SAF}L_{RR}S) model are: first row image (lambda _2=50), (k=32), second row image (lambda _2=500), (k=10), third row (lambda _2=100), (k=30) and the last row image (lambda _2=500), (k=30). These images were taken from the kaggle datasets33 and are available publicly.
Figure 10 demonstrates the comparison among ARKFCM method34 with different waited filters (ARKFCM(_a), ARKFCM(_m) and ARKFCM(_w)), Cai et al. method14 and the anticipated (M_{SAF}L_{RR}S) model. The first column of this figure displays the given images, the second, third and the fourth columns display all segmented outcomes with ARKFCM(_a), ARKFCM(_m) and ARKFCM(_w), respectively, whereas the fifth and last columns show segmented results of Cai and (M_{SAF}L_{RR}S) methods, respectively. Figure 10 shows comparison of those methods on three brain MR images which suffer from intensity inhomogeneity. As ARKFCM framework consists of three methods i.e. average, mean and weighted, therefore we tested each image of Fig. 10 we compare our model with all of them. In fact, this can be well understood from all outcomes and these figures that all the three methods of ARKFCM framework are unable to segment the given brain MR images by missing significant details in it. Similarly, in the fifth column of this figure we can see that Cai model14 also is unable to segment these images properly. In contrast, the segmented results of (M_{SAF}L_{RR}S) model (last column) capture all the significant details present in these MR images of brain.
The Jaccard similarity measure, and the CPU time (which is measured in seconds) of the proposed and other state-of-the-art models are given away in Table 1. The statistics deliberate that the anticipated model is better than other state-of-the-art model, including the well-known Chan-Vese11 and Cai et al.14 models, in terms of higher Jaccard similarity (JS) ratio and lower CPU or computational time. This should be noted that variations in the JS values and CPU times are, in fact, due to the heterogeneity of the images and their different characteristics.
Segmenting a synthetic image, using the LBF approach, Cai et al.14 method, and the suggested (M_{SAF}L_{RR}S) approach. The first column represents the original image and its noisy version, while the second column characterizes the result of the LBF method. Furthermore, the third and forth columns represent the segmented outcomes of the Cai et al.14 and proposed model, respectively. Furthermore, the parameters used for (M_{SAF}L_{RR}S) model are (d) (lambda _2=100), (k=10) and for (h) and (l) are (lambda _2=200), (k=20). The images were taken from Ref.33 and are publicly available.
Segmentation results on three dissimilar MR images of the brain. Note that, these images were taken from the kaggle repository33. The first column out of the six columns represents original images, while second, third and fourth columns represent segmentation outcomes of ARKFCM(_a), ARKFCM(_m) and ARKFCM(_w), respectively. Similarly, the fifth and the last column demonstrates segmented results of the Cai et al.14 and (M_{SAF}L_{RR}S) models, respectively. In all these three experiments (three rows), the parameters used for our proposed model are (lambda _2=30) and (k=30).
Besides the Jaccard similarity coefficient other quantitative can also be implemented to assess and quantify the performance of the proposed image segmentation approach. The Sørensen–Dice similarity values are normalized and given with in the limit of [0, 1]. The greater value for the Dice shows relatively superior segmentation outcomes and vice versa. Very similar to the Jaccard similarity coefficient, the Sørensen–Dice similarity index value is computed using the following Eq. (20):
where X represents the ground truth, Y denotes the obtained image and D shows the Sørensen–Dice similarity between Y and X. Table 2 shows the Sørensen–Dice coefficients values for comparison purposes of the anticipated approach with other competitory techniques, i.e., Cai et al.14, LBF16, Chan-Vese11, and ARKFCM34. This should be noted that these results were attained using numerical simulations on 10 various images that were in fact appropriate for interactive segmentation using a particular predefined ground truth value. Note that the value is assumed to be comprising means of the ground truth. This can be determined that the Cai et al. model generated comparatively superior outcomes as equated to the LBF and ARKFCM models. However, for a richly noisy or depleted intensity image the anticipated approach loses the statistics. From the outcomes, as shown in Table 2, this is clear that the suggested approach performs superior to the other competing techniques. The Chan-Vese11 performs relatively better than the Cai et al.14 model but the Sørensen–Dice similarity value of the suggested approach is still higher than the Chan-Vese model11. In fact, the higher values of the deviations show that the model perform completely different over various images.
In this paper, we developed a new hybrid variational approach for image segmentation, restoration, and filtering. The planned approach is especially tailored for those images that suffer with the intensity inhomogeneity, noise or brightness in the background. For this purpose we took advantage of utilizing dual filter formulation and fuzzy membership function. We tested our proposed model on a variety of images including real life images of plane and public data set of fingerprints images and observed that the proposed model can tackle weak fingerprint images very well. We also proved that our model is more accurate and fast as compared to available models used for the same task. We observed in empirical evaluation and through the obtained outcomes that the proposed model (M_{SAF}L_{RR}S) segments the edges of various images very efficiently and accurately. we compared our proposed model with other well known and latest models used for image segmentation and proved that our method perform better than these methods. Our future goal is to modify the proposed model for severe blurry and foggy images as these are difficult to segment and restore with the existing methods.
One of the most fundamental fields in real-world applications and medical imaging is selective image segmentation. In the future, we will work on and deliver a strong selective segmentation approach grounded on the concept of local spatial distance; and simultaneously utilising a dual-level set variational formulation model. A comparable approach should attempt to divide all image items through a single level set function (aka. global) and the chosen item through a dissimilar level set function (aka. local). Additionally, in the future, the use of a mix and amalgamation of marker distance function, and local spatial distance. Nevertheless, we will continue investigating the edge detection, and active contour without edges should also be examined in parallel. In the existence of outliers and noise, especially the Gaussian noise, outliers should be recognised and separated during the pre-processing of denoising. This should be kept in mind that suitable constraints should be put forward and assumed to the segmentation framework in order to ensure and guarantee proper and acceptable picture segmentation. In the future, we will use more appropriate, robust, and appropriate approaches for removing outliers’ and criteria, in particular, integrated with and backed by a well-designed hypothesis in a variational model for precise and acceptable image restoration and segmentation.
The datasets generated and/or analysed during the current study are publicly available in the kaggle repository, and can be accessed at [https://www.kaggle.com/datasets/mnavaidd/image-segmentation-dataset]. Moreover, various images used within the experimental work are publicly available online. All the codes used for this method will be provided for research purposes if requested by researchers.
Rudin, L. I., Osher, S. & Fatemi, E. Nonlinear total variation based noise removal algorithms. Physica D 60, 259–268. https://doi.org/10.1016/0167-2789(92)90242-F (1992).
ADS MathSciNet Article MATH Google Scholar
Strong, D. & Chan, T. Edge-preserving and scale-dependent properties of total variation regularization. Inverse Prob. 19, S165 (2003).
ADS MathSciNet Article Google Scholar
Chan, T., Marquina, A. & Mulet, P. High-order total variation-based image restoration. SIAM J. Sci. Comput. 22, 503–516. https://doi.org/10.1137/S1064827598344169 (2000).
MathSciNet Article MATH Google Scholar
Lysaker, M., Lundervold, A. & Tai, X.-C. Noise removal using fourth-order partial differential equation with applications to medical magnetic resonance images in space and time. IEEE Trans. Image Process. 12, 1579–1590. https://doi.org/10.1109/TIP.2003.819229 (2003).
ADS Article PubMed MATH Google Scholar
Cai, J.-F., Chan, R. H. & Shen, Z. A framelet-based image inpainting algorithm. Appl. Comput. Harmon. Anal. 24, 131–149. https://doi.org/10.1016/j.acha.2007.10.002 (2008) (Special Issue on Mathematical Imaging – Part II).
MathSciNet Article MATH Google Scholar
Goldstein, T. & Osher, S. The split bregman method for l1-regularized problems. SIAM J. Imag. Sci. 2, 323–343. https://doi.org/10.1137/080725891 (2009).
MathSciNet Article MATH Google Scholar
Brito-Loeza, C. & Chen, K. On high-order denoising models and fast algorithms for vector-valued images. IEEE Trans. Image Process. 19, 1518–1527. https://doi.org/10.1109/TIP.2010.2042655 (2010).
ADS MathSciNet Article PubMed MATH Google Scholar
Chambolle, A. & Pock, T. A first-order primal-dual algorithm for convex problems with applications to imaging. J. Math. Imaging Vis. 40, 120–145. https://doi.org/10.1007/s10851-010-0251-1 (2011).
MathSciNet Article MATH Google Scholar
Csiszar, I. Why least squares and maximum entropy? An axiomatic approach to inference for linear inverse problems. Ann. Stat. 19, 2032–2066. https://doi.org/10.1214/aos/1176348385 (1991).
MathSciNet Article MATH Google Scholar
Nikolova, M. A variational approach to remove outliers and impulse noise. J. Math. Imaging Vis. 20, 99–120. https://doi.org/10.1023/B:JMIV.0000011326.88682.e5 (2004).
MathSciNet Article MATH Google Scholar
Chan, T. F. & Vese, L. A. Active contours without edges. IEEE Trans. Image Process. 10, 266–277. https://doi.org/10.1109/83.902291 (2001).
ADS CAS Article PubMed MATH Google Scholar
Mumford, D. & Shah, J. Optimal approximation by piecewise smooth functions and associated variational problems. Commun. Pure Appl. Math. 42, 577–685 (1989).
MathSciNet Article Google Scholar
Gout, C., Le Guyader, C. & Vese, L. Segmentation under geometrical conditions using geodesic active contours and interpolation using level set methods. Numer. Algorithms 39, 155–173. https://doi.org/10.1007/s11075-004-3627-8 (2005).
ADS MathSciNet Article MATH Google Scholar
Cai, X. Variational image segmentation model coupled with image restoration achievements. Pattern Recogn. 48, 2029–2042. https://doi.org/10.1016/j.patcog.2015.01.008 (2015).
ADS Article MATH Google Scholar
Mabood, L., Ali, H., Badshah, N., Chen, K. & Khan, G. A. Active contours textural and inhomogeneous object extraction. Pattern Recogn. 55, 87–99. https://doi.org/10.1016/j.patcog.2016.01.021 (2016).
ADS Article Google Scholar
Li, C., Kao, C., Gore, J. C. & Ding, Z. Implicit active contours driven by local binary fitting energy. In 2007 IEEE Conference on Computer Vision and Pattern Recognition, 1–7. https://doi.org/10.1109/CVPR.2007.383014 (2007).
Wang, X.-F., Huang, D.-S. & Xu, H. An efficient local Chan-Yese model for image segmentation. Pattern Recogn. 43, 603–618. https://doi.org/10.1016/j.patcog.2009.08.002 (2010).
ADS Article MATH Google Scholar
Zhang, K., Song, H. & Zhang, L. Active contours driven by local image fitting energy. Pattern Recogn. 43, 1199–1206. https://doi.org/10.1016/j.patcog.2009.10.010 (2010).
ADS Article MATH Google Scholar
Dong, F., Chen, Z. & Wang, J. A new level set method for inhomogeneous image segmentation. Image Vis. Comput. 31, 809–822. https://doi.org/10.1016/j.imavis.2013.08.003 (2013).
Article Google Scholar
Wang, X.-F., Min, H. & Zhang, Y.-G. Multi-scale local region based level set method for image segmentation in the presence of intensity inhomogeneity. Neurocomputing 151, 1086–1098. https://doi.org/10.1016/j.neucom.2014.01.079 (2015).
Article Google Scholar
Kumar, M., Thurow, K., Stoll, N. & Stoll, R. Fuzzy filtering: A mathematical theory and applications in life science. In Fuzzy Systems (ed. Azar, A. T.) (IntechOpen, 2010).
Google Scholar
Yoo, J., Ho Lee, S. & Kwak, N. Image restoration by estimating frequency distribution of local patches. CoRR. http://arXiv.org/abs/1805.09097 (2018).
Caselles, V., Kimmel, R. & Sapiro, G. Geodesic active contours. Int. J. Comput. Vis. 22, 61–79. https://doi.org/10.1023/A:1007979827043 (1997).
Article MATH Google Scholar
Guyader, C. L. & Gout, C. Geodesic active contour under geometrical conditions: Theory and 3d applications. Numer. Algorithms 48, 105–133 (2008).
ADS MathSciNet Article Google Scholar
He, Y., Hussaini, M. Y., Ma, J., Shafei, B. & Steidl, G. A new fuzzy c-means method with total variation regularization for segmentation of images with noisy and incomplete data. Pattern Recogn. 45, 3463–3471. https://doi.org/10.1016/j.patcog.2012.03.009 (2012) (Best Papers of Iberian Conference on Pattern Recognition and Image Analysis (IbPRIA '2011)).
ADS Article MATH Google Scholar
Li, F., Ng, M., Zeng, T. & Shen, C. A multiphase image segmentation method based on fuzzy region competition. SIAM J. Image Sci. 3, 277–299. https://doi.org/10.1137/080736752 (2010).
MathSciNet Article MATH Google Scholar
Feng, C., Zhao, D. & Huang, M. Image segmentation and bias correction using local inhomogeneous intensity clustering (linc): A region-based level set method. Neurocomputing 219, 107–129 (2017).
Article Google Scholar
Sethian, J. A. Level Set Methods and Fast Marching Methods: Evolving Interfaces in Computational Geometry, Fluid Mechanics, Computer Vision, and Materials Sciences (Cambridge University Press, 2002).
MATH Google Scholar
Osher, S. & Fedkiw, R. Level Set Methods and Dynamic Implicit Surfaces (Springer, 2003).
Book Google Scholar
Li, C. et al. A variational level set approach to segmentation and bias correction of images with intensity inhomogeneity. Med. Image Comput. Comput. Assist. Interv. 11, 1083–1091 (2008).
PubMed PubMed Central Google Scholar
Feng, C., Zhao, D. & Huang, M. Image segmentation using cuda accelerated non-local means denoising and bias correction embedded fuzzy c-means (bcefcm). Signal Process. 122, 164–189 (2016).
Article Google Scholar
Feng, C., Zhao, D. & Huang, M. Segmentation of longitudinal brain mr images using bias correction embedded fuzzy c-means with non-locally spatio-temporal regularization. J. Vis. Commun. Image Represent. 38, 517–529 (2016).
Article Google Scholar
Kaggle Datasets. https://www.kaggle.com/datasets. (Accessed 18 February 2022).
Elazab, A. et al. Segmentation of brain tissues from magnetic resonance images using adaptively regularized kernel-based fuzzy (c)-means clustering. Comput. Math. Methods Med.https://doi.org/10.1155/2015/485495 (2015).
Article PubMed PubMed Central MATH Google Scholar
Cai, X. & Steidl, G. Multiclass segmentation by iterated rof thresholding. In Energy Minimization Methods in Computer Vision and Pattern Recognition (eds Heyden, A. et al.) 237–250 (Springer, 2013).
Chapter Google Scholar
Vairy, M. & Venkatesh, Y. Deblurring Gaussian blur using a wavelet array transform. Pattern Recogn. 28, 965–976. https://doi.org/10.1016/0031-3203(94)00146-D (1995).
ADS Article Google Scholar
Shan, Q., Jia, J. & Agarwala, A. High-quality motion deblurring from a single image. ACM Trans. Graph 27, 73:1-73:10. https://doi.org/10.1145/1360612.1360672 (2008).
Article Google Scholar
Yun, S. & Woo, H. Linearized proximal alternating minimization algorithm for motion deblurring by nonlocal regularization. Pattern Recogn. 44, 1312–1326. https://doi.org/10.1016/j.patcog.2010.12.013 (2011).
ADS Article MATH Google Scholar
Minaee, S. et al. Image segmentation using deep learning: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence (2021).
Zhang, M., Dong, B. & Li, Q. Deep active contour network for medical image segmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention, 321–331 (Springer, 2020).
Hatamizadeh, A. et al. Deep active lesion segmentation. In International Workshop on Machine Learning in Medical Imaging, 98–105 (Springer, 2019).
Zhang, Y. & Chung, A. Deep supervision with additional labels for retinal vessel segmentation task. In International Conference on Medical Image Computing and Computer-Assisted Intervention, 83–91 (Springer, 2018).
Ghosh, S., Das, N., Das, I. & Maulik, U. Understanding deep learning techniques for image segmentation. ACM Comput. Surveys (CSUR) 52, 1–35 (2019).
Article Google Scholar
Fan, L., Zhang, F., Fan, H. & Zhang, C. Brief review of image denoising techniques. Vis. Comput. Ind. Biomed. Art 2, 1–12 (2019).
CAS Article Google Scholar
Pires, R. G., Santos, D. F., Santos, C. F., Santana, M. C. & Papa, J. P. Image denoising using attention-residual convolutional neural networks. In 2020 33rd SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI), 101–107 (IEEE, 2020).
Wang, G. et al. Interactive medical image segmentation using deep learning with image-specific fine tuning. IEEE Trans. Med. Imaging 37, 1562–1573 (2018).
Article Google Scholar
Boyd, S., Parikh, N., Chu, E., Peleato, B. & Eckstein, J. Distributed optimization and statistical learning via the alternating direction method of multipliers. Found. Trends Mach. Learn. 3, 1–122. https://doi.org/10.1561/2200000016 (2011).
Article MATH Google Scholar
Goldstein, T. & Osher, S. The split Bregman method for l1-regularized problems. SIAM J. Image Sci. 2, 323–343. https://doi.org/10.1137/080725891 (2009).
MathSciNet Article MATH Google Scholar
Pock, T., Chambolle, A., Cremers, D. & Bischof, H. A convex relaxation approach for computing minimal partitions. In 2009 IEEE Conference on Computer Vision and Pattern Recognition, 810–817. https://doi.org/10.1109/CVPR.2009.5206604 (2009).
Yuan, J., Bae, E., Tai, X.-C. & Boykov, Y. A continuous max-flow approach to potts model. In Computer Vision—ECCV 2010 (eds Daniilidis, K. et al.) 379–392 (Springer, 2010).
Chapter Google Scholar
Rahman, A. et al. A selective segmentation model using dual-level set functions and local spatial distance. IEEE Access 10, 22344–22358 (2022).
Article Google Scholar
Download references
This research is supported by the University of Peshawar and the Abdul Wali Khan University Mardan (AWKUM), Pakistan.
These authors contributed equally: Noor Badshah, Muhammad Zakarya, Ayaz Ali Khan and Muhammad Haleem.
Department of Mathematics, University of Peshawar, Peshawar, Pakistan
Lutful Mabood & Haider Ali
Department of Basic Sciences, University of Engineering and Technology, Peshawar, Pakistan
Noor Badshah
Department of Computer Science, Abdul Wali Khan University, Mardan, Pakistan
Muhammad Zakarya & Aftab Ahmed
Department of Computer Science, University of Lakki Marwat, Khyber Pakhtunkhwa, Pakistan
Ayaz Ali Khan
Biomedical Engineering Department, Bahcesehir University, Besiktas, Istanbul, Turkey
Lavdie Rada
Department of Computer Science, Kardan University, Kabul, Afghanistan
Muhammad Haleem
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
L.M.: Research, Methodology, Writing—Original Draft; N.B.: Conceptualization, Methodology, Software, Writing—Review & Editing; H.A.: Conceptualization, Visualization, Validation, Investigation; M.Z.: Visualization, Data Curation, Proofreading; A.A.: Visualization, Validation, Proofreading; A.A.K.: Writing—Review & Editing, Revisions; L.R.: Visualization, Writing—Review & Editing; M.H.: Writing—Revised Draft, Data Curation.
Correspondence to Muhammad Haleem.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Mabood, L., Badshah, N., Ali, H. et al. Multi-scale-average-filter-assisted level set segmentation model with local region restoration achievements. Sci Rep 12, 15949 (2022). https://doi.org/10.1038/s41598-022-19893-z
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-19893-z
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
'Eager and ready': CCC brings on new staff and student support – berkeleyhighjacket.com
Erika Pistoni has replaced Mary Jacobs in the CCC
Jahan Ingraham
Pistoni is the admin assistant in the College and Career Center (CCC) this year, a job filled by Mary Jacobs for many years prior.
“I book the appointments with the college representatives, manage their meetings with students, in person or online,” Pistoni said. She communicates with seniors and parents, and will begin talking to juniors this spring.
This is her first year working in this position at the CCC, but Pistoni has been working in the district since 2020, before COVID-19 lockdowns.
“I took a test to enter the district … and they hired me as a sub on Friday, and on Monday we were on lockdown,” Pistoni said. “That weekend I was so happy, like ‘Yay! I’ve been hired, I got some work,’ And then no one worked for two years.”
As school reopened, she worked in various positions such as substitute teaching and as the student record assistant, maintaining academic records. When the admin assistant position became available, she was happy to take it.
“I really enjoy being with students, so I really pushed to be here,” Pistoni said. She hopes to carry on the tradition and add smiling faces to the photo wall Jacobs left up at her old desk.
Erika Pistoni has replaced Mary Jacobs in the CCC.
Sakima Williams helps students find jobs and internships in the CCC.
Jahan Ingraham
For Williams, a new CCC counselor, “(the) goal is to be a part of the new legacy of Berkeley (High) … and the amazing talent that’s about to come out of here.” Williams also teaches two classes at BHS: Pop Culture: Inventing Hip Hop, and Audio Recording and Engineering.
In the CCC, Williams helps students find jobs and internships. He has a list up on the center’s wall with potential jobs available to teens in Berkeley. In his role, Williams emails students about opportunities for jobs and internships.
“(I) also bring industry into the school to do pop ups and signups, for everything from auto mechanics to bioengineering … both within the city of Berkeley as well as the Greater Bay Area,” Williams said.
Previously, Williams worked in audio recording and music production. He also worked as a counselor for alcohol and drug prevention, and as an advocate for African American male achievement.
“I’ve been in youth development for the last 25 years,” Williams said. “I dodged bullets in Rodeo at the refinery in the housing projects, all the way to playing guitar for kids … in an after school program.”
Williams struggled as a student with learning disabilities, and the experience of “educational trauma” gave him “insight into what young people go through.” As a teacher, he understands the purpose of his work at BHS is to be a part of the school’s legacy.
“Now is the time that I step back and facilitate the opportunity for young people to do it themselves,” Williams said. “Because let’s face it, music, revolution, change, everything comes from y’all, not us.”
Sakima Williams helps students find jobs and internships in the CCC.
Maya Martinez is an ‘intensive advisor’ for about 100 BHS students.
Jahan Ingraham
Martinez is the new College Adviser Fellow in the CCC.
“I don’t have a whole repertoire of being a counselor, but I’m still eager and ready to get things going,” Martinez said. Working with the Destination College Advising Corps and the Early Academic Outreach Program (DCAC EAOP), she assists low income, first generation, and underrepresented students with support to reach college goals.
Martinez supports about 100 students with “intensive advising,” making sure they meet requirements and preparing them for the college process.
“I remember when I was about to go to college I was like, ‘Out of sight, out of mind,’” Martinez said. “We want to familiarize people with (the college process) … so that way it’s not like a scary monster under the bed.”
Martinez is a graduate from UC Davis, interested in education. At Davis, Martinez was a teacher’s assistant and planned to become a teacher. That door is still open; College Adviser Fellows in the DCAC EAOP program stay for a few years.
Martinez has big goals to equip her students with as much expertise as possible.
“I would love to see a lot of people … not only being enrolled but actually getting some fruit from (the program),” said Martinez. “Because let’s be real, college can affect you for the rest of your life. I want to inspire kids, and make sure that y’all are equipped in the best way possible.”
To learn more about the program visit @EAOPDCACBHS.
Maya Martinez is an ‘intensive advisor’ for about 100 BHS students.
By clicking above, you agree to Disqus’ privacy policy and our comment policy
We provide the opportunity to comment in order to foster a healthy debating environment and reserve the right to reject comments that stray away from that objective. Read our full policy →
The Voice of the Students
© 1912-2022 Berkeley High Jacket
Website by Jerome Paulos ’22
SCSD Employee Completes Master Teacher of Economics Program – MageeNews.com
Please note that this post contains affiliate links and any sales made through such links will reward MageeNews.com a small commission – at no extra cost to you.
Simpson County School District (SCSD) employee Kenyatta Durr is now recognized as a leader in Mississippi’s economic development, thanks to her completion of the Master Teacher of Economics certificate program from the Mississippi Council on Economic Education (MCEE).
Durr, a health informatics instructor at Simpson County Technical Center (SCTC), successfully completed the Master of Teacher of Economics program this summer from the MCEE. Completion of the program attests to her knowledge of basic and advanced-level economic concepts and her ability to help students achieve economic and financial literacy. Durr was accepted into the program, along with 18 other participants from various backgrounds and industries.
“Mrs. Durr is an outstanding addition to the SCTC instructional staff. The fact that she is willing to improve in her craft by completing this program is a testament to that” said SCTC Director George Huffman. “With the completion of the Master Teacher of Economics program she will be able to provide more in-depth instruction related to economics and how it impacts the healthcare profession.”
The Master Teacher of Economics Program is an intense 16-week professional development program offered in collaboration with the MCEE and the Mississippi Department of Education to provide professional development to teachers in economics and financial education. The Department of Education confers the Master Teacher of Economics designation.
Having completed other certificate programs from the MCEE before, Durr said she pursued the economics certificate to give herself an extra skill she could bring to the classroom to enhance classroom instruction. Since completion of the program, Durr said she has been able to add things she learned from the program to her curriculum and is impressed at her students’ response to the material.
“Since going through the program and having access to some of the resources that come with completing the course, I have found it beneficial and that it has helped prepare me for the classroom,” said Durr. “Simply engaging my students in conversation and asking thought-provoking questions has allowed us to talk about particular issues and how it impacts the economy, the types of jobs available and the amount of money they can make.”
According to Durr, her students are developing an understanding of relationships between things such as supply and demand and how disastrous events can impact the local, state and national economy.
“Once they start learning some of those economic skills, they are able to make relationships from what is happening in our economy, based on their studies,” she said. “Now, when they come into the classroom, they are telling me about things that they see on the news and asking questions about how it will impact our economy.”
This winter and spring, Durr’s students will have an opportunity to showcase their new skills and knowledge in the Mississippi National Economics Challenge – a competition to engage students more fully in their learning while assessing how well they are learning economic concepts. The students will compete with other students from across the state to apply their economics knowledge and answer rigorous questions on microeconomics, macroeconomics, international economics and current events for a chance to compete on the national level.
A native of Simpson County, Durr has worked with the SCSD for two years. Prior to her current role as health informatics instructor SCTC, she served as an instructor of business marketing and finance and of international business for Magee High School.
MageeNews.com is an online news source serving Simpson and surrounding counties as well as the State of Mississippi.
This site uses Akismet to reduce spam. Learn how your comment data is processed.
Political Forum
Magee Board of Alderman 9/20/22
Albert Turnage for Chancery Judge, Post 1
How to watch Northwestern vs. Miami (OH): Live stream, TV channel, start time for Saturday's NCAA Football game – CBS Sports
Play Now
Football Pick’em
Play Now
College Football Pick’em
Start or Sit (NFC Home Games)
Talking the business of sports
Miami (OH) @ Northwestern
Current Records: Miami (OH) 1-2; Northwestern 1-2
The Northwestern Wildcats have the luxury of staying home another week and will welcome the Miami (OH) RedHawks at 7:30 p.m. ET Sept. 24 at Ryan Field. Both teams took a loss in their last game, so they’ll have plenty of motivation to get the ‘W.’
It was all tied up 14-14 at halftime, but the Wildcats were not quite the Southern Illinois Salukis’ equal in the second half when they met last week. Northwestern came up short against SIU, falling 31-24. If the result catches you off guard, it should: Northwestern was far and away the favorite. One thing holding Northwestern back was the mediocre play of QB Ryan Hilinski, who did not have his best game: despite two touchdowns, he threw two interceptions and fumbled the ball once with only 4.95 yards per passing attempt.
Meanwhile, Miami (OH) scored first but ultimately less than the Cincinnati Bearcats in their game last week. The RedHawks’ painful 38-17 defeat to Cincinnati might stick with them for a while. Miami (OH) was surely aware of their 24-point disadvantage in the point spread, but the foreknowledge did nothing to prevent the result. No one had a standout game offensively for Miami (OH), but they got scores from WR Mac Hippenhammer and TE Jack Coldiron.
The losses bumped both teams down to an identical 1-2. A pair of defensive numbers to keep in mind before kickoff: Northwestern enters the contest having picked the ball off four times, good for 21st in the nation. Miami (OH) is completely their equal: they also come into the matchup with four interceptions.
The Wildcats are a big 7.5-point favorite against the RedHawks, according to the latest college football odds.
The oddsmakers had a good feel for the line for this one, as the game opened with the Wildcats as a 6.5-point favorite.
Over/Under: -109
See college football picks for every single game, including this one, from SportsLine’s advanced computer model. Get picks now.
This is the first time these teams have played each other within the last seven years.
© 2004-2022 CBS Interactive. All Rights Reserved.
CBS Sports is a registered trademark of CBS Broadcasting Inc. Commissioner.com is a registered trademark of CBS Interactive Inc.
Images by Getty Images and US Presswire
Prometheus pays $188M for San Bruno apartment complex – The Real Deal
For full functionality of this site it is necessary to enable JavaScript. Please Allow Javascript and reload this page.
Prometheus Real Estate Group has acquired over 400 garden-style apartments in San Bruno for $188 million, its second multifamily acquisition in the Bay Area in as many months and the sector’s second-priciest trade in San Mateo County this year.
The San Mateo-based apartment landlord and manager paid about $431,000 per unit for the 435-unit Crystal Springs Terrace at 2000 Crystal Springs Road, according to public records. The sellers were Pivnicka Properties and Gerson Bakar & Associates, and the deal closed Sept. 19, documents filed with the San Mateo County Clerk-Recorder’s Office show.
Crystal Springs Terrace is sandwiched between Interstate 280 and Crystal Springs Road and sits in the bayside foothills of the Santa Cruz Mountains. Completed in 1974, it contains a mix of studio, one- and two-bedroom apartments spread across more than two dozen buildings, according to Prometheus’ website and title service records. Units range from 522 to 965 square feet, with monthly lease rates between $2,225 and $3,795 a month, according to Prometheus’ site.
The priciest one-bed, one-bath apartment available for rent costs $2,588 a month, below the citywide average of $3,180 for a one-bed unit, Zumper data show. However, the property’s priciest studio and two-bed units cost between $178 and $450 more a month to rent than the citywide average, according to Zumper data.
Representatives for Prometheus and Filbert Partners didn’t respond to requests for comment. The latter’s Stephen LoPresti has overseen the acquisition, finance, and asset management of Gerson Bakar’s real estate portfolio for the past two decades, according to Filbert’s website.
Only one other multifamily property in San Mateo County– Bell South City in South San Francisco–has sold for more than Crystal Springs Terrace this year according to commercial real estate analytics firm Reonomy. North Carolina’s Bell Partners paid $206 million in May to acquire the 260-unit complex from Sares Regis Group of Northern California, which works out to about $793,000 a unit.
That’s well above what Prometheus paid on a per-unit basis for Crystal Springs Terrace, and the difference in age between the two properties may have something to do with it. Sares Regis delivered Bell South City to the market in 2019, while Crystal Springs was built nearly 50 years ago. While the latter has been renovated at least once since then, according to reviews of the property on Apartments.com, it’s unclear when that happened or how extensive the renovations were.
The deal for Crystal Springs Terrace closed a month after Prometheus acquired another apartment complex in the Bay Area, paying $52 million for a 94-unit building in Cupertino. The apartment owner and manager also spent $48 million last year to purchase the site of an office building in Los Altos that has the city’s OK to be redeveloped into 196 condos and townhomes.
Prometheus was already one of the Bay Area’s largest apartment landlords and managers before closing those deals, overseeing nearly 10,000 units in Silicon Valley alone, according to the Silicon Valley Business Journal. The company’s portfolio encompasses 13,000 apartments in the Bay Area, Oregon and Washington, with another 2,600 units in the pipeline, according to its website.
All rights reserved © 2022 The Real Deal is a registered Trademark of Korangy Publishing Inc.
The Deal’s newsletters give you the latest scoops, fresh headlines, marketing data, and things to know within the industry.
By clicking Subscribe you agree to our Privacy Policy.
Round-up top news and topics for each of the following cities
Select the newsletter you’d like to receive below
Select the newsletter you’d like to receive below
Select the newsletter you’d like to receive below
BandLab – Content Strategy Manager (Remote) (US) – Music Business Worldwide
Content Strategy Manager, BandLab Sounds
Los Angeles, California, United States Marketing. Full time
BandLab is a social music platform that empowers everyone to be a creator. With a global community of over 50 million musicians and fans, BandLab combines music-making tools like the world’s first cross-platform collaborative audio workstation with social features like video sharing, messaging, and discovery.
As Content Strategy Manager, BandLab Sounds, you will be responsible for developing catalogue and content strategies based on BandLab Sounds’ business objectives and our users’ needs. In this role, you will translate strategy into planning, execution and optimization to ensure BandLab Sounds is the most inspiring and accessible library of royalty-free samples, loops and one-shots on the internet.
Diversity and Inclusion
BandLab Technologies commits to fostering diversity and inclusion in all that we do. Regardless of race, gender identity, sexual orientation, disability, religion or any other status protected by law, we offer applicants the equal opportunity to join, belong and flourish in our organisation.
Music Business Jobs is the most effective way for the global music industry to find the best candidates for their open positions.
In addition to being seen by over 300,000 monthly unique readers online, our listings are also emailed direct to more than 50,000 professional subscribers every week – across the US, UK, Europe, Asia-Pacific and beyond.
Click through below for details on how to submit a Music Business Jobs listing – and learn why our advertisers ‘strongly recommend’ us to their peers…
South Carolina vs. Charlotte: How to watch online, live stream info, game time, TV channel – CBS Sports
Play Now
Football Pick’em
Play Now
College Football Pick’em
Start or Sit (NFC Home Games)
Talking the business of sports
Charlotte @ South Carolina
Current Records: Charlotte 1-3; South Carolina 1-2
The Charlotte 49ers are staring down a pretty large 23.5-point disadvantage in the spread for Saturday’s game. They are staying on the road Saturday to face off against the South Carolina Gamecocks at 7:30 p.m. ET Sept. 24 at Williams-Brice Stadium. The 49ers will be strutting in after a victory while USC will be stumbling in from a defeat.
Last week, Charlotte won a matchup that couldn’t have been any closer, slipping by the Georgia State Panthers 42-41. Charlotte’s QB Chris Reynolds was on fire, passing for five TDs and 401 yards on 43 attempts.
Meanwhile, the day started off rough for USC last week, and it ended that way, too. It’s hard to picture a worse loss than the 48-7 bruising that they suffered against the Georgia Bulldogs. USC was in a tough position by the half, with the score already sitting at 24 to nothing. QB Spencer Rattler had a memorable game, but not in the way you want to be remembered: besides failing to produce a single touchdown, he threw two interceptions with only 4.72 yards per passing attempt.
The 49ers are now 1-3 while the Gamecocks sit at 1-2. Two offensive stats to keep in the back of your head while watching: Charlotte ranks seventh in the nation when it comes to passing touchdowns, with ten on the season. USC has displayed some offensive firepower of their own as they enter the contest with 272.3 passing yards per game on average, good for 38th best in the nation.
The Gamecocks are a big 23.5-point favorite against the 49ers, according to the latest college football odds.
The oddsmakers had a good feel for the line for this one, as the game opened with the Gamecocks as a 23-point favorite.
Over/Under: -110
See college football picks for every single game, including this one, from SportsLine’s advanced computer model. Get picks now.
This is the first time these teams have played each other within the last seven years.
© 2004-2022 CBS Interactive. All Rights Reserved.
CBS Sports is a registered trademark of CBS Broadcasting Inc. Commissioner.com is a registered trademark of CBS Interactive Inc.
Images by Getty Images and US Presswire
Virat Kohli was chase master, Babar Azam has mastered the art: Shoaib Akhtar – The Indian Express
Shoaib Akhtar on Thursday heaped praise on Pakistan skipper Babar Azam after he scored a brilliant century against England in the second T20I and compared his style of chasing the big totals to former India captain Virat Kohli.
“Virat Kohli was chase master, Babar Azam has mastered that art,” said Akhtar on his Youtube channel.
Babar made an unbeaten 110 in his second T20 century off 66 balls as Pakistan reached 203-0 with three balls to spare to level the seven-match series. Mohammad Rizwan also scored 88.
Babar and Rizwan charged against England with aggressive batting as the visitors couldn’t hold on to their chances to break the stand.
“This is what Pakistan requires — the run and strike rates,” said Akhtar.
Babar came into Thursday’s game on the back of only 98 runs in his previous seven T20s that included 68 runs in the six Asia Cup matches in the United Arab Emirates and a knock of 30 in the opening game of the series on Tuesday.
“Both the openers, especially Babar Azam. He showed once again, why is no 1 player of the world. When he scores quickly, he makes life easier for Rizwan. They complemented each other,” added Akhtar.
Babar and Rizwan held the previous record of Pakistan’s 197-run, opening-wicket stand, which they made against South Africa at Centurion in 2021.
Akhter maintained the Azam and Rizwan must follow the same template in the upcomming match so that the middle order can have some breathing space.
“Strike Rate is very important. Yes, you can’t maintained the this strike rate in every match, but make sure that you leave the middle order some space so that they don’t have to chase with require run rate going above 13,” said Akhtar.
Delhi High Court dismisses plea of Jamia contract employee seeking regularisation of services
69 apartments + retail topped out at 5109 W. Adams Boulevard – Urbanize LA
Los Angeles
One block west of the intersection of Adams Boulevard and La Brea Avenue, developer CIM Group has completed the wooden frame of one of its largest projects yet in West Adams.
The project, now taking shape at 5109 W. Adams Boulevard, consists of a six-story building that will feature 69 studio, one-, and two-bedroom dwellings at completion. Plans also call for 5,689 square feet of ground-floor retail and restaurant space and underground parking for 63 vehicles.
Construction at 5109 W Adams BoulevardUrbanize LA
Project entitlements include Transit Oriented Communities incentives, permitting greater height and density than allowed by zoning rules in exchange for setting aside seven residential units as deed-restricted affordable housing.
Designed by Studio One Eleven, 5109 Adams is depicted in renderings with a contemporary podium-type building with offset windows on its upper floors Planned on-site amenities include amenity decks at the second and sixth floors, as well as a rooftop open space.
5109 West Adams BoulevardGoogle Street View
Prior to initiating the current plan for 5109 Adams, CIM Group sought to entitle the site as well as an adjoining lot for the construction of a five-story, 99-unit apartment building. The entitlement application for that project was withdrawn in 2018, and building permits now indicate that a four-story, 78-room hotel is slated for the neighboring property. That project, if completed, would be CIM’s second hotel in the West Adams neighborhood, following the 43-room Alsace at 5170 W. Adams Boulevard.
CIM is developing dozens of properties in the West Adams and Jefferson Park neighborhoods, including office buildings and similar apartment complexes. The company recently cleared a commercial building two blocks west at 5201 W. Adams Boulevard for the construction of another mixed-use apartment complex.
Copyright © Urbanize Media LLC 2022
The UB Interview: Boney James Talks New Album 'Detour' | Stream 'Detour' – UrbanBridgez.com

Jazz/R&B sax player Boney James has released his 18th studio album, “Detour” via Concord Records.
Featuring the breezy current hit single “Coastin” with 5-time Grammy winner Lalah Hathaway
on vocals, the new 10-song collection is a subtle stylistic turn that builds on the commercial success of 2020’s “Solid,” his highest charting release ever on the pop charts, peaking in the Top 10 on the Billboard 200.
Other standouts on “Detour” include the track “Bring It Back”, which released mid-summer and features critically acclaimed composer, conductor, orchestrator, arranger, and trumpeter, Dontae Winslow; and the album’s title track “Detour,” a winding, playful duet between sax and piano that features some of Boney’s most adventurous playing to date.
“Detour” was recorded largely at Boney’s L.A. home studio (live drums and percussion were tracked at venerable Sunset Sound). The album was produced by Boney with co-production from in-demand writer/producer, Jairus “J-Mo” Mozee among others. Additional musicians on the album are keyboardist Tim Carmon, longtime drummer Omari Williams and bassist Alex Al, renowned percussionist Lenny Castro and guitarists Big Mike Hart and Paul Jackson Jr.
Boney James will continue on the road for select dates across the U.S. which will take him well into 2023.
During his 30-year career, Boney James has released a staggering 18 studio albums – including a pair of Christmas records -with 1995’s Seduction, 1997’s “Sweet Thing” and 1999’s “Body Language” all certified gold. In 2015, Boney’s “futuresoul” spent 11 weeks atop the Billboard Contemporary Jazz Chart and was the best-selling Contemporary Jazz Record of 2015. Boney has earned four Grammy nominations, two NAACP Image Award nominations and a Soul Train Award, while a dozen of his albums have landed at #1 on Billboard’s Contemporary Jazz Album chart.
His collaborators in the past include a Who’s Who of R&B and soul: Raheem DeVaughn, Faith Evans, George Benson, George Duke, Dwele, Al Jarreau, Philip Bailey, Anthony Hamilton, Jaheim, Eric Benét, Dave Hollister, Stokley Williams, Kenny Lattimore and Angie Stone.
“Detour” will be released on vinyl on January 13th, 2023. This will mark the first time in Boney James’ career that he will release an album on vinyl.
In celebration of “Detour,” Aries from UB spoke to Boney James about his 18th album and how his collaboration with legendary Lalah Hathaway came about.
Boney also gets candid on lessons he’s learned in the music business and tells us what we can expect from a Boney James show and more!
Upcoming Tour Dates;
Fri, OCT 14 – Rehoboth Beach Jazz Festival – Lewes, DE
Thu, OCT 27 – Florida Theatre – Jacksonville, FL
Sun, NOV 6 – Cancun Jazz Fest – Cancún, Mexico
Fri, NOV 11 – Jazz Legacy Foundation Fest – Hampton, VA
Thu, NOV 17 – Birchmere – Alexandria, VA
Fri, NOV 18 – Rams Head Early Show – Annapolis, MD
Fri, NOV 18 – Rams Head Late Show – Annapolis, MD
Sun, JAN 15, 2023 – Punta Cana Jazz Festival – Punta Cana, Dominican Republic
Mon, JAN 23, 2023 – Smooth Jazz Cruise 2023 – Miami Gardens, FL
Mon, JAN 30, 2023 – Smooth Jazz Cruise 2023 – Miami Gardens, FL
Sat, FEB 18, 2023 – The Palladium at the Center for the Performing Arts – Carmel, IN

Enter your email to subscribe to UB and receive post and interview updates once daily, by email.
[…] Credit: Source link […]…
[…] Source link […]…
[…] The UB Story of Legendary Gerald LeVert | Keith Sw…
[…] Source link […]…
[…] Credit: Source link […]…
Don't have an account?
You must be logged in to post a comment.
New Mexico State vs. Hawaii: How to watch online, live stream info, game time, TV channel – CBS Sports
Play Now
Football Pick’em
Play Now
College Football Pick’em
Start or Sit (NFC Home Games)
Talking the business of sports
Hawaii @ New Mexico State
Current Records: Hawaii 1-3; New Mexico State 0-4
The New Mexico State Aggies will be returning home after a three-game road trip. They will take on the Hawaii Rainbow Warriors at 8 p.m. ET Saturday at Aggie Memorial Stadium. New Mexico State lost both of their matches to Hawaii last season on scores of 21-41 and 34-48, so they’re hoping to turn the tables this season.
The Aggies played a game they are hoping to forget as they lost a 66-7 blowout to the Wisconsin Badgers last week. New Mexico State was in a tough position by the half, with the score already sitting at 35 to nothing. Their only offensive touchdown came from RB Ahmonte Watkins.
Meanwhile, Hawaii netted a 24-14 victory over the Duquesne Dukes last week.
New Mexico State is the favorite in this one, with an expected 4.5-point margin of victory. True fans might be the only ones betting on them, currently 0-3 ATS, to cover the spread.
Hawaii’s win lifted them to 1-3 while New Mexico State’s loss dropped them down to 0-4. We’ll see if the Rainbow Warriors can repeat their recent success or if the Aggies bounce back and reverse their fortune.
The Aggies are a 4.5-point favorite against the Rainbow Warriors, according to the latest college football odds.
The line has drifted a bit towards the Aggies, as the game opened with the Aggies as a 2-point favorite.
Over/Under: -110
See college football picks for every single game, including this one, from SportsLine’s advanced computer model. Get picks now.
Hawaii have won both of the games they’ve played against New Mexico State in the last eight years.
© 2004-2022 CBS Interactive. All Rights Reserved.
CBS Sports is a registered trademark of CBS Broadcasting Inc. Commissioner.com is a registered trademark of CBS Interactive Inc.
Images by Getty Images and US Presswire
Master of the Beat: Ben Bakson – outsmartmagazine.com
Ben Bakson grew up an only child in a small village in south Germany, watching movies, listening to music and enjoying long days on the lake with his family. At 18, he moved to Berlin to study Economics and after graduating, he landed a marketing job at a large firm. He found the 9 to 5 life wasn’t for that and began dabbling in music production.
His music style today is commercial with hard, emotional, and energetic beats. He regards vocals as vital, referring to them as the soul of every song. Music fans seem to agree as Ben Bakson has been playing to packed dancefloors every weekend for three months, spinning parties in Sao Paulo, San Francisco, Mykonos, Washington DC and NYC.
We caught up with the master of the beat for a phone chat before he travels to Houston to spin at Rebar on Saturday, Sept 24th.
What’s the worst part of doing these kind of interviews?
Ben Bakson: Never knowing what kind of questions are going to be asked! (Laughing) You need to have the right answer ready immediately. But I am quite quick-witted, to be honest, therefore it has not been an issue so far.
You seem like the quiet-type; someone who prefers to keep his private life to himself.
Interesting perception, because I think most people think the opposite. They assume that I am outgoing because I’m a DJ that celebrates my life quiet publicly on social media. They don’t realize all of that is part of my job. I do try to keep my private life as private as possible, especially my love life.
It seems to me like a lot of deejays choose to speak their minds through their music sets.
I definitely agree, and I absolutely do that. Music is the one common language we humans have. For me, it is therapy. I can channel my emotions through music and I love when I get feedback from fans that my music helped them to get through a difficult moment. That makes me very happy.
Would you say your music as tribal?
I’m not a fan of labeling my music as a particular genre because I honestly think my music style crosses several genres. I feel the same way about being forced to label my sexuality. Am I gay simply because I like sex with men? Sex is a very small part of who I am as a person and I think everyone should have sex with the person he loves and finds attractive. I don’t think it needs an explanation, label or commenting. I’m convinced that in future the whole idea of straight, gay and whatever we do in bed won’t matter because, in the end, what counts is that we are all human and equal.
Is that why your dancefloors attract all genders and sexualities?
I try to make my dancefloor as diverse as I can. I have a lot of straight friends that I invite to my events and they love my music as much as they gay men and women. My big goal is for my sound is to be loved by everyone. I would like to play all floors. The big hurdle for me is that most of the big straight festivals won`t book any of the big DJs of the LGBT world.
What will it take to blur the lines between gay and straight?
We need party producers to create events that attract straight and gay people and then book DJs from both worlds that will unite all and create a unique musical journey.
Tell us about your new song, “You Got Me.”
The track was written by the talented singer OMEO. It’s about a relationship with a toxic person and how important it is to free yourself from them.
Is it a song you can relate with?
Most definitely. I have been single for about a year now which has been the longest time in my life so far. Honestly, I have been very happy, sharing my time with amazing friends and my beloved dog Macho. I may have met a special person but I’m not putting any stress on it or making any long-term plans.
Who is your celeb crush?
That’s easy: JLo’s hairdresser, Chris Appleton.
If you could have any magic power, what would it be?
Sometimes I think I already have too many magic powers. Like reading people’s minds and knowing what will happen next. I have a sixth sense! I also think music is my magic power. I can bring people together, make them feel happy on the dancefloor and even help them feel better at home through my podcasts. That’s pretty powerful.
Besides music, do you have a secret talent?
I think I do, but I can’t talk about it here.
What is your idea of perfect happiness?
That moment when I realize how grateful I am with all that I have.
What is your greatest fear?
The music stopping during my set. It has happened twice so far, both due to technical issues that were out of my control.
What is your most treasured possession?
My dog, Macho.
If you could change one thing about yourself, what would it be?
There are things I’d like to change and you can be sure I’m working on changing them already.
What is the trait you most deplore in others?
Envy. I don’t really understand people who try to sabotage others. I prefer to be happy for them and support their success. It motivates me to work harder to be more successful myself.
Which living person do you most admire?
My parents. I am very thankful for what they gave me in my childhood. Education, lots of love and empathy; without them I would not be where I am now.
What do you most dislike about your appearance?
I think I am ok. We all need to learn to relax a bit about being too self-critical and just be happy the way we are.
What makes Macho the greatest dog in the world?
You have to meet him and you will understand immediately. The cutest, most loving, protecting, most intelligent being I have ever met.
Follow Ben on Instagram @ ben_bakson.
Special master calls for help in Trump Mar-a-Lago documents fight – POLITICO
Legal
The former president also gets a deadline to produce any proof of tampering during the FBI search last month.
A car passes in front of former President Donald Trump's Mar-a-Lago resort on Feb. 11, 2022, in Palm Beach, Fla. | Joe Raedle/Getty Images
By Josh Gerstein
Link Copied
The outside expert tapped to sort through former President Donald Trump’s legal claims over documents seized from his Florida home last month is calling for backup.
U.S. District Court Judge Raymond Dearie, a New York City-based jurist named by a federal judge in Florida to act as a so-called special master in the review of more than 11,000 documents the FBI confiscated, proposed on Thursday that former Magistrate Judge James Orenstein help with the process.
“The undersigned has determined that the efficient administration of the Special Master’s duties requires the assistance of the Honorable James Orenstein,” Dearie’s proposed plan for the document review said. The order said Orenstein “has experience with complex case management, privilege review, warrant procedures, and other matters that may arise in the course of the Special Master’s duties.”
Orenstein spent 16 years as a federal magistrate in the same Brooklyn courthouse where Dearie sits. Orenstein drew attention several years ago for his role in what was semi-sarcastically dubbed “the magistrates’ revolt” — rulings from a smattering of federal magistrate judges across the country questioning government tactics in warrant applications seeking electronic data.
In 2016, Orenstein issued a controversial ruling rejecting prosecutors’ arguments that a two-century-old federal law gave the government the right to command Apple to assist in unlocking an iPhone used by an alleged drug dealer. The judge’s pro-privacy stance in that matter may have led the secretive Foreign Intelligence Surveillance Court to appoint him to a list of approved friends of the court who provide their perspective on surveillance requests.
Dearie proposed that the former magistrate, who has a top secret clearance, be paid $500 an hour for work on the Trump documents case. The federal judge who named Dearie as special master, Florida-based Aileen Cannon, has previously ruled that Trump must assume all expenses related to the review.
Dearie said in the Thursday order that he would not be compensated for his work because he’s still being paid his judge’s salary, but that expenses such as Orenstein’s fees must be paid in full and on time or Trump could face court sanctions.
While the seven-page proposal the special master issued on Thursday bears Dearie’s name and a form of electronic signature, metadata attached to the document indicates that Orenstein — who retired from the court in 2020 — was involved in drafting it. Dearie’s chambers did not immediately respond to a request for comment on whether Orenstein helped prepare the proposal to involve him in the case.
Dearie’s plan, still subject to comment from the parties, gives Trump a deadline of Sept. 30 to indicate whether he disputes the FBI’s list of what was seized in the search. Last month, Trump suggested that such tampering was a possibility given the FBI’s refusal to allow a lawyer for Trump to witness the search.
“The FBI and others from the Federal Government would not let anyone, including my lawyers, be anywhere near the areas that were rummaged and otherwise looked at during the raid on Mar-a-Lago,” Trump wrote on his Truth Social media site last month. “Everyone was asked to leave the premises, they wanted to be left alone, without any witnesses to see what they were doing, taking or, hopefully not, ‘planting.’”
Trump has not offered any evidence of such tampering by the FBI, but Dearie’s plan says that if Trump’s team wants to press such a claim, someone will need to submit a statement under penalty of perjury detailing the grounds for it.
The proposed plan calls for the review of the more than 11,000 documents seized from Mar-a-Lago to take place in four batches. The first involves a small set of information the Justice Department has already flagged as potentially subject to attorney-client privilege. Trump’s team has weekly deadlines stretching through Oct. 14 to identify allegedly privileged material and explain why. Prosecutors have parallel deadlines through Oct. 21, with Dearie scheduled to submit his recommendations to Cannon by Oct. 31
As of now, roughly 100 documents with various national security classification markings that were seized during the search will not be part of the special master process. Cannon ruled that they should be and rebuffed a request by prosecutors to carve those records out.
But on Wednesday night, a federal appeals court sided with the Justice Department, ruling that the government’s broad authority over classified information made it a mistake for Cannon, a Trump appointee, to include the alleged secrets in the wider review and for her to put them off-limits to criminal investigators in the meantime.
Trump’s lawyers have yet to indicate whether they plan to ask the Supreme Court to step in and restore Cannon’s original order.
It took little time for the 11th U.S. Circuit Court of Appeals ruling against Trump to boomerang and help prompt a legal loss for one of his allies: MyPillow founder Mike Lindell.
Last week, the FBI used a search warrant to seize Lindell’s mobile phone as he waited at a Hardee’s drive-through in Mankato, Minn. Lindell filed suit on Tuesday in federal court in Minnesota, seeking a temporary restraining order requiring the government to stop accessing his phone and ultimately to return it.
However, a judge denied the restraining order on Thursday, citing, among other things, the 11th Circuit ruling against Trump a day earlier.
U.S. District Court Judge Eric Tostrud, a Trump appointee, said that Lindell’s right to get his phone back “is not obvious, and that’s understating things.” The judge’s order didn’t rule out the possibility that Lindell might be entitled to get his phone back after both sides have the chance to develop their legal arguments in court filings.
Lindell’s phone appears to have been seized in connection with an ongoing investigation into alleged tampering with voting machines in Colorado last year.
CORRECTION: An earlier version of this article misstated the focus of the investigation that led to the seizure of Mike Lindell’s phone.
Link Copied
© 2022 POLITICO LLC
Howl's Moving Castle – Studio Ghibli Fest 2022 – Creative Loafing Tampa
Rated PG
The next date Howl’s Moving Castle – Studio Ghibli Fest 2022 is playing is Sunday, September 25 .
(Howl’s Moving Castle – Studio Ghibli Fest 2022 is not showing in any area theaters on Saturday, September 24.)
Subscribe now to get the latest news delivered right to your inbox.
Tampa now has a magic mushroom dispensary
By Colin Wolf
Family of Andrew Joseph III awarded $15 million in lawsuit against Hillsborough Sheriff, calls win a blow to qualified immunity
By Justin Garcia
Florida Cracker Fish Company will open out of former Ballyhoo Grill location in Citrus Park
By Kyla Fields
St. Pete’s Independent Bar is closing, new brewery in Seminole Heights, and more in local foodie news
By Kyla Fields
View more issues
Read our sister publications
633 N. Franklin Street, Suite 735
Tampa, FL 33602
A high-resolution flux-matrix model describes the spread of diseases in a spatial network and the effect of mitigation strategies | Scientific Reports – Nature.com
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 15946 (2022)
Metrics details
Propagation of an epidemic across a spatial network of communities is described by a variant of the SIR model accompanied by an intercommunity infectivity matrix. This matrix is estimated from fluxes between communities, obtained from cell-phone tracking data recorded in the USA between March 2020 and February 2021. We apply this model to the SARS-CoV-2 pandemic by fitting just one global parameter representing the frequency of interaction between individuals. We find that the predicted infections agree reasonably well with the reported cases. We clearly see the effect of “shelter-in-place” policies introduced at the onset of the pandemic. Interestingly, a model with uniform transmission rates produces similar results, suggesting that the epidemic transmission was deeply influenced by air travel. We then study the effect of alternative mitigation policies, in particular restricting long-range travel. We find that this policy is successful in decreasing the epidemic size and slowing down the spread, but less effective than the shelter-in-place policy. This policy can result in a pulled wave of infections. We express its velocity and characterize the shape of the traveling front as a function of the epidemiological parameters. Finally, we discuss a policy of selectively constraining travel based on an edge-betweenness criterion.
When plague hit Florence in August 1630, the Florentine authorities made a number of high-stakes decisions which proved highly effective1. One of the reasons the Florence Sanitá could organize this response was the ample time they had, forewarned as they were by the Milanese authorities in November 1629. Today’s public-health authorities work under much more compressed timescales, as evidenced by the SARS-CoV-2 pandemic. Long-distance travel radically changes the dynamics of spreading, which raises a number of questions about the spatial dynamics of transmission in modern times. Epidemic outbreaks in the last two decades have provided the scientific community with a wealth of material to study these questions, going beyond the classic Susceptible-Infected-Recovered (SIR) theory with perfect mixing2,3,4,5,6. Several studies have shown how the total epidemic size can be affected by factors such as inhomogeneity in transmission rates7,8,9,10,11,12,13,14 or in the mode of transmission15,16. Classically, motion of individuals was taken into account by introducing diffusion terms in the standard SIR equations, allowing the emergence of spatio-temporal patterns17,18,19,20. Recently, in the context of the SARS-CoV-2 pandemic, such approaches have been especially valuable in order to study the effect of containment policies such as lockdown and quarantine21,22. These models are, however, limited in that they do not, in principle, take long-distance air travel into account. Several works have, therefore, considered disease spread in a network, typically constructed from air-traffic data19,23,24, where edges can connect locations separated by large geographical distances. This approach can lead to very accurate predictions at the country scale25 but predictions at finer scales remain challenging. Another study considering human mobility emphasized how spatial variation in public-health infrastructure reflected on epidemiological parameters can affect the dynamics of spread to different countries26. Data-based studies of epidemic spread and the impact of social distancing through the analysis of social-network structure have also been very informative16,27,28,29.
A recent paper by Chang et al.30 obtained a model for the spatio-temporal spread of a disease at a high spatial resolution by using extensive mobile tracking information to identify physical interactions between individuals. Chang et al. showed that the actual spread of the SARS-CoV-2 epidemic can be well explained from the mobility data of individuals. The model relied on the simulation of interactions among individuals on a bipartite graph where nodes, representing locations at a very fine spatial resolution, are divided in two sets: Census Block Groups (residential areas) and Points Of Interest (non-residential), each of them having its own transmission rate. The model was fitted to reproduce known reported cases of COVID-19 in 10 metropolitan areas, and could then be used to make short-term predictions about the spreading or study the effect of different mitigation strategies.
Here we take an approach similar to that of Chang et al., using mobility data to calibrate a model for disease spread. However, we investigate this propagation at the scale of a large country, the USA, rather than metropolitan areas. Specifically, we introduce a spatial SIR epidemiological model in which effective transmission rates between (N={2^{10}}) communities are computed from mobility data of individuals belonging to these communities. We show that this model captures well the spread of the SARS-CoV-2 epidemic. Remarkably, we find that a simple model consisting of an interaction frequency dropping under the effect of lockdown, and of a single flux matrix encoding the travel of individuals, faithfully reproduces the reported cases of COVID-19 both globally and locally in each community. Strikingly, the SARS-CoV-2 epidemic spreads in a delocalized fashion, infecting distant communities very quickly. Moreover, an even simpler model with uniform transmission rates between the communities gives results very close to the model based on mobility data, emphasizing the prevalent role air travel played in the spread of the SARS-CoV-2 epidemic. We then study how interventions that change travel patterns can localize epidemics. In particular, we investigate the hypothetical effect of a policy preventing long-distance travel. In addition to “flattening” the curve, spreading through nearest-neighbor interactions creates traveling waves, which we characterize both analytically and numerically. These results allow us to discuss which interventions are more effective, limiting short-range contacts (a lockdown), or limiting long-range trips (a quarantine). We also propose an alternative mitigation strategy based on an edge-betweenness criterion.
We consider a metapopulation model of N communities numbered 1, 2,…, N. Denoting by (S_a), (I_a) and (R_a) the numbers of susceptible, infected and recovered individuals in community a, the standard SIR equations read:
where (beta _{ab}) is the transmission rate from infected individuals in community b to susceptible individuals in community a and (gamma) is the recovery rate, assumed to be the same for all communities. Diagonal elements of the infectivity matrix ([beta _{ab}]) describe intracommunity infections, while off-diagonal elements describe inter-community infections (Fig. 1a,b). We also introduce the local epidemic sizes (T_a = I_a + R_a). The total population in each community (M_a) is constant through time,
The model in Eq. (1) has been extensively studied31,32,33,34,35. We show in the Supplementary Information (SI) that the dynamics can be reduced to an ODE of just one N-vector variable, and the endemic equilibrium can be obtained by solving a transcendental equation involving the infectivity matrix ([beta _{ab}]).
Model for the spreading of SARS-CoV-2 in a network of communities in the USA. (a) One community i interacting with three other communities j, k and l, with the transmission rates (beta _{ij}), (beta _{ik}) and (beta _{il}) respectively. (b) Infectivity matrix. (c) (N=2^{10}) communities in the USA. Each community aggregates a number of Census Block Groups (CBGs).
In order to estimate the infectivity matrix, we used mobility data compiled by SafeGraph36, tracking the location of about 20 million USA cell phones between March 2020 and February 2021. The locations consist of more than 200,000 Census Block Groups (CBGs). Each cell phone is assigned for physical residence the location where it spent the most time, and daily visits to other locations are recorded. For computational purposes, we coarse-grained the physical locations into (N = {2^{10}}) communities, which are shown in Fig. 1c. Let (f_{ab}) be the number of individuals from community a visiting community b per unit of time. We will assume that
The variation in susceptible individuals in community a due to new infections during the time interval (Delta t) has the form:
where (beta) is the disease-specific transmission rate when a susceptible individual has contact with an infected individual. There are three kinds of infection to consider: (i) the infected person and the infector belong to the same community, (ii) an infected person visits a neighboring community and infects a resident of this community, and (iii) a susceptible person visits a neighboring community and gets infected by one of its residents. We will neglect the rarer “tourist to tourist” infections, when an infected person visits a neighboring community and infects there a visitor from yet another community. The term (mathrm {Pr}left( text {meeting an infected individual}right)) can therefore be evaluated as a function of the pseudo-flux matrix ([f_{ab}]) for each of the three aforementioned cases (SI). After summation of the three contributions, we obtain:
where p is the frequency with which an individual is having contact with an other individual of its community, and is assumed to be the same for all communities. Equation (5) determines the infectivity matrix ([beta _{ab}]) from Eq. (1) up to a proportionality constant, namely (beta p).
In order to assess the validity of the infectivity matrix based on mobility data, Eq. (5), we confronted the model’s predictions against COVID-19 case numbers reported in the USA by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University37. Specifically, we fitted the daily (beta p(t)) values so as to minimize the sum of squared errors between values predicted by the model and values reported by the CSSE (Material and methods). The fitted (beta p(t)) values show a steep decay during the month of March 2020, followed by a plateau lasting until February 2021 (Fig. 2b).
Model based on SafeGraph mobility data or a uniform infectivity matrix. (a) The model is fitted to COVID-19 confirmed cases in the USA. There is one fitting parameter per day. (b) Fitting parameters obtained. The shape suggests a simplified model with two limiting values before and after lockdown. (c) The simplified models obtained reproduce the spread of the SARS-CoV-2 epidemic in the community network. A direct comparison between the local epidemic sizes predicted by the models and the reported values can be found in Fig. 5. See also Movies S1 and S2.
The time variations of (beta p(t)) shown in Fig. 2b imply a drastic decrease in the interaction frequency among individuals across all communities. This result reflects the effect of the shelter-in-place policies that were implemented at the beginning of the SARS-CoV-2 pandemic in many USA states. We can use this observation to estimate the effect of these policies quantitatively: the interaction frequency among individuals, namely p, is about 5 times smaller in the plateau following shelter-in-place policies than it was at the onset of the SARS-CoV-2 pandemic. Following this observation, we defined a simplified model, in which the fit with reported cases was carried out while enforcing a softplus shape for (beta p(t)) (Fig. 2b and Material and methods). This simplified model reproduced the average progression of the epidemic (Fig. 2a), but it didn’t capture the three oscillations visible in the number of new cases. We conclude that these oscillations in the number of new cases were mostly driven by a similar oscillatory pattern in the interaction frequency as seen in Fig. 2b. In Fig. 2c, we show for a few dates an overlay of the number of new COVID-19 cases in each community according to the official reports and as predicted by our simplified model. Despite non-negligeable variance, we find that the model reproduces to a satisfying degree the real dynamics. This can also be seen through a direct comparison of the model predictions with the real reports of new cases, for all dates combined, as shown in Fig. 5a. This result suggests that the infectivity matrix constructed from mobility data (Eq. 5) is a reasonable approximation of the “true” infectivity matrix.
The previous results suggest that the decrease in new COVID-19 cases was mostly driven by a country-wide reduction in the interaction frequency among individuals. Here we investigate the hypothetical effect of an alternative policy, namely a travel restriction while keeping unchanged the interaction frequency among individuals. Specifically, we modified the infectivity matrix so that communities separated by a physical distance larger than a prescribed cutoff do not interact: (beta _{ij} = 0) if (d_{ij} > d_c) (Fig. 3a). We seeded the infection in a community belonging to the state of Washington and simulated the spread of the epidemics using a fixed interaction frequency ((beta p(0)) from the simplified model). As expected, we observed a reduction in the number of daily new cases dT (we define the local variables (dT_a(t) = T_a(t)-T_a(t-1))) when the cutoff distance (d_c) decreased, illustrating the “curve-flattening” effect that was targeted by travel restriction policies (Fig. 3b). In this idealized scenario with a single seed for the infection, the epidemic propagates as a traveling wave from the west coast to the east coast (Fig. 3c). However, as can be seen by comparing Fig. 2a to Fig. 3b, travel restriction policies are not as efficient as lockdown policies to decrease the spread of an epidemic.
Limiting long-distance travels without local lockdown. (a) Only transmission rates for communities separated by a distance (d_{ij} < d_c) are retained. (b) Daily new infections for increasing values of the cutoff distance. (c) Spatial visualization of the daily new infections using a cutoff (d_c = {200};{text {km}}). See also Movie S3.
The daily infectivity matrices constructed from the SafeGraph mobility data can be viewed as elements of a random-matrix ensemble. Remarkably, matrix elements seem to be distributed according to the law (beta _{ij} = langle beta _{ij} rangle e^{xi }), where (langle beta _{ij} rangle) is the mean infectivity matrix and (xi equiv N(0,1)) is a centered reduced Gaussian variable (Fig. S1). The probability density of the eigenvalues is also shown in Fig. S1. A connection can be made with random matrix theory (RMT)38,39,40, initially introduced by E. Wigner to model the spectra of the nuclei of heavy atoms, where the interactions between many nucleons are assumed to be drawn from a random ensemble. In RMT, random matrices are classified according to their symmetry, or according to their corresponding level statistics (namely, the probability density of the spacing between consecutive eigenvalues) that exhibit different degrees of level repulsion38. In particular, the Wigner-Dyson (WD) statistics is typical of the Gaussian Orthogonal Ensemble (in which eigenvalues “interact”), while the Poisson statistics is typical of a matrix with independent eigenvalues. The level-spacing statistics of the infectivity matrices is shown in Fig. 4a. We found that they were time-independent. Yet interestingly, the level statistics interpolates between the WD statistics and the Poisson statistics41. As shown earlier, removing links between communities according to their geographical distance results in a slower spread and a smaller epidemic size. Concomitantly, the level spacing distribution converges toward a Poisson statistics (Fig. 4c). The crossover from WD (entropy (S=0.7169)) to Poisson ((S=1)) distribution as links between communities are successively removed suggests an isolation policy that can lead to an effective reduction of the epidemic size (SI). As alternative mitigation strategies, we choose to induce the transition toward a Poisson distribution by decimating links between communities according to their “nominal” distance or the “edge-betweenness” centrality42,43,44 (Fig. 4b). Figure 4d shows how the level spacing distribution converges toward a Poisson statistics when the nominal distance threshold is lowered. We find the edge-betweenness centrality to be more efficient in decreasing the epidemic size. This is because decimation according to edge-betweenness first targets links with the largest transmission rates. One could also consider a moderate policy: instead of eliminating links completely, one could impose constraints on the flux of individuals that are allowed to commute via central pre-determined links.
Random-matrix properties and mitigation strategies. (a) Level spacing distribution (unfolded38,45,46,47) of infectivity matrices between 2020-03-01 and 2021-02-15 (pooled by 7-day windows). The solid lines denote the Wigner-Dyson and the Poisson statistics. The level distribution is shown in Fig. S1. (b) The epidemic size (Omega /Omega _0) versus the relative number of cuts (N_c/N/(N-1)) for the 3 mitigation methods: distance cutoff, bandwidth, and edge-betweenness centrality. (Omega _0) is the epidemic size for (N_c=0). Level-spacing entropy S and epidemic size as a function of (c) the cutoff distance (d_c) in the range 170-12,000 km, and (d) the half-bandwidth (1{le }B<1023).
Although SafeGraph mobility data36 provide a realistic picture of people movement between communities, one might ask what is the sensitivity of the results to the infectivity matrix derived from the mobility fluxes. We thus considered a model with uniform transmission rates among communities, namely (beta _{ab} = beta), which suppresses spatial effects. In particular, this model leads to the natural variables (nu _a) (see SI) to be uniform: (nu _a(t) = nu ^text {UN}(t)). Surprisingly, carrying out the same fit to the reported cases (Fig. 1a,b) was only marginally inferior to the fit carried out with the SafeGraph mobility data (see Fig. 5b). By contrast, carrying out the same fit with the infectivity matrix derived from SafeGraph mobility data but with long-range interactions turned down resulted in a significantly different dynamics (see Fig. 5c). There are several implications of this result. (1) Although changes in the structure of the infectivity matrix can lead to drastically different dynamics (Figs. 3, 4 and 5c), it appears that the infectivity matrix derived from SafeGraph fluxes falls in the same “universality class” as the uniform model. This suggests that long-range movements (e.g. air traffic) played a prevalent role in the spread of SARS-CoV-2. (2) Our choice to reduce the complexity of the model to only one fitting parameter might not be adequate to discriminate between models falling into the same universality class. Instead of fitting only one scalar p(t) at each time point, we also investigated the possibility of fitting the (N^2) transmission rates (beta _{ab}(t)) minimizing the errors with reported cases (see Material and methods and Fig. S2). Although this latter approach is clearly prone to overfitting, it shows that there exists a parametrization of the model which reproduces very closely reported cases. We anticipate that the proper number of fitting parameters lies in between those two extreme scenarios.
Comparison of the (fitted) model predictions with reported values using (a) the infectivity matrix derived from SafeGraph fluxes (same model as Fig. 2), (b) an infectivity matrix with uniform transmission rates, and (c) the infectivity matrix derived from SafeGraph fluxes truncated to geographical distances (d_c < {400} ; text{km}). Local epidemic sizes (T_a) are shown. One symbol is associated to one given community and one given day.
The results from the previous sections suggest the apparition of a wavefront when transmissions are short range. To investigate this phenomenon, we consider a simplified model of communities lying on a two-dimensional square lattice. Each community index is replaced by coordinates ((i,j) in llbracket 1, 2^n rrbracket times llbracket 1, 2^m rrbracket). In particular, we consider that only individuals from neighboring communities interact together. We rewrite Eq. (1):
where (alpha) (respectively (beta)) is the intra-community (resp. inter-community) transmission rate. After rescaling the time and space variables, and defining the rescaled recovery rate (tilde{gamma } = gamma / (a beta )) where (a = 4 + alpha /beta), we look for wave solutions in the continuum limit by introducing the shape functions (S(x,y,t) = g(x – tilde{v} t)) and (I(x,y,t) = h(x – tilde{v} t)), where (tilde{v} = v / sqrt{a} beta) represents the velocity of the wave in rescaled time and space (see SI). The shape functions satisfy the ODE:
We find that the velocity of the traveling wave is bounded from below (SI):
which is in agreement with previous reports18,48 and with results from marginal-stability analysis21,49,50. Interestingly, by an independant argument, we also established (see SI) that (tilde{v} le tilde{v}_c). Therefore the traveling wave must move at velocity (tilde{v} = tilde{v}_c). This indicates that the SIR dynamics in Eq. (6) falls into the Fisher-Kolmogorov-Petrovsky-Piscunov universality class, resulting in pulled waves51,52,53,54. Although we have taken the continuum limit of a nearest-neighbor model, this analysis is also valid for any finite-range infection matrix with the appropriate rescaling of variables.
We performed simulations of the dynamics given by Eq. (6) on a square lattice with a varying aspect ratio (Fig. 6a) and seeding the infection at one site on the west boundary. As expected, there is a front of new infections, moving from west to east as time progresses. A timelapse of a traveling front of infected individuals with (alpha = beta = gamma = {0.1}) is shown in Fig. 6b. The wave position increases asymptotically linearly with time, but the velocity (tilde{v}) of the wave varies with (tilde{gamma }) (Fig. 6c). The profiles obtained are in agreement with the shape functions obtained by solving Eq. (7), as shown in Fig. 6d.
Existence of a wave with nearest-neighbors-only interactions. (a) Equation (6) is solved on a square lattice of (2^n times 2^m) sites. (b) Traveling front of infected individuals moving along the x direction (left to right). We took (alpha = beta = gamma = {0.1}), so that the rescaled recovery rate is (tilde{gamma } = gamma / (a beta ) = {0.2}). (c) The dynamics simulated for different values of (beta). The position m(t) ((S(t, m(t))=(1 + S_infty )/2)) of the wave is asymptotically linear in time, hence a constant velocity (tilde{v} = sqrt{1-tilde{gamma }}). (d) The simulated wave profiles (symbols) are in agreement with the profiles predicted by Eq. (7) (solid lines). The corresponding S,I,R profiles are shown in Fig. S3. A simulation with (tilde{gamma } = {0.2}) is shown in Movie S4.
Biological systems are inherently complex and it can be a challenge to characterize them by a small number of parameters. In the case of epidemics, the huge number of parameters (for example, intercommunity spread involves (N^2) transmission rates for N communities) can obscure the salient mechanisms and make it difficult for policy makers to find efficient interventions. Reducing the dimension of a model, when possible, is therefore of great value. In this article, we have proposed a spatial SIR model with an infectivity matrix based on the local travel patterns.
Despite its simplicity (there is only one global parameter to be adjusted), we find that our model is able to capture the spatial spread of the SARS-CoV-2 epidemic, with a delocalized multicenter spreading caused by long-distance travelers bringing infection into far regions, then becoming secondary centers of infection. The time evolution of the interaction frequency reflects the shelter-in-place policies that were implemented by various USA states in the early stages of the SARS-CoV-2 pandemic. In fact, the rich diversity of human responses could be summarized by a simplified model for the interaction frequency, whose asymptotes represent the values “before lockdown” and “after lockdown”, with few assumptions about the infection process.
If a model with relatively few parameters describes the observations, one can assume that it also describes the situation when these parameters are changed by an intervention. Therefore, we suggest such a model could be used as the basis for efficient policies—or at least reliably to estimate the consequences of adopting policies. As an example, we have shown the hypothetical effect of an alternative to the shelter-in-place policy in the case of the SARS-CoV-2 pandemic. Specifically, we have investigated a travel restriction policy in which individuals can only move within an area of fixed radius centered around their residence. By contrast to the lockdown policy, we find that infections spread through a well-defined wave front, traveling with a certain velocity. This scenario might be preferable since it gives time to communities and public health infrastructures to prepare for the onset of the epidemics, while still “flattening the curve” of new infections (Fig. 3b). We have also provided both an analytical and numerical analysis elucidating the mechanism of formation of a wavefront. In particular, we give the velocity and the shape of the wavefront for an epidemic spreading through nearest neighbors interactions.
During the course of our research, an epidemiological study bearing similarities with our approach was published55. In that study, the authors developed a county-resolved metapopulation model describing the spread of SARS-CoV-2 in the USA, informed with mobility data from SafeGraph. Instead of calibrating the model by fitting the country-wide number of reported cases as in the present study, the authors fitted their model to reproduce reported cases of COVID-19 on a per-county basis. Furthermore, the model required fitting many time-dependent parameters on a per-county basis, including per-county transmission rates, making the fitting procedure very high dimensional. Finally, the dynamics of the disease spread introduced differs from Eq. (1) since S, I and R compartments were introduced for each commute channel (i rightarrow j) rather than for each community. Altogether, the complexity of that model makes it less amenable to analytical study.
In conclusion, we used a simple model of intercommunity spread of an infectious disease to show the transition between different regimes of epidemic progression. Because of the complexity of the infection process (e.g., variations in individuals’ responses, mutations, etc.) and of human behavior, we are still far from a global forecasting system able to predict the spread of different infectious agents throughout the world. As with weather forecasting, observables must be measured in real time in order to inform complex models to yield short-term forecasts. One salient feature in our approach is showing how such widespread measurements (namely, the mobility data) can be integrated in a model describing the spread of an infectious agent, and showing what types of predictions can be obtained. It is a step toward more predictive epidemiology models, grounded in measurable quantities.
The mobility data was obtained from SafeGraph, a company that aggregates anonymized location data from numerous applications in order to provide insights about physical places, via the SafeGraph Community. To enhance privacy, SafeGraph excludes census block group (CBG) information if fewer than two devices visited an establishment in a month from a given census block group. In this manuscript, we use data extracted from the “Social Distancing Metrics” dataset36, with dates between February 2020 and February 2021. SafeGraph has stopped sharing mobility data under this format, however the same data can be acessed under a different format through the Neighborhood Patterns dataset56. However the matrices of fluxes between our coarse-grained communities can be found in the ‘data’ folder in our github repository. We obtained the coarse-grained communities by running a K-means clustering algorithm to group the 220,333 CBGs into ({2^{10}}) communities. We used the implementation from Scikit Learn. We then ran a hierarchical clustering algorithm to re-index communities so that communities close in space had close indices, as shown in Fig. 1. We used the linkage function from SciPy. We then computed for every day the “flux matrix” where each entry (f_{ab}) represents the number of cell phone whose residence CBG belongs to community a which visited a CBG belonging to community b. The average flux matrix was constructed by averaging all the daily flux matrices. We used population counts for CBGs in agreement with the United States Census Bureau’s and available in the SafeGraph Open Census Data57 (file “cbg_b01.csv”, column “B01001e1”). We checked that the population counts from the United States Census Bureau were approximately proportional to the residential mobile-phone counts, therefore validating mobile tracking as a proxy for actual population.
In order to fit our model, we used USA cases of COVID-19 reported by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University37 which can be accessed at the GitHub repository github.com/CSSEGISandData/COVID-19. New cases of COVID-19 reported by the CSSE were mapped to the closest mobility-data-derived community based on their latitude and longitude. We therefore obtained the time evolution of SARS-CoV-2 infections through the mobility-data derived communities. Integer absolute new cases were converted into relative population fractions using the community population counts obtained from the United States Census Bureau.
Here we consider the SIR dynamics (see Supplementary Information):
We also define the chi-square at each time t:
where (M_a) is the population at location a and ((1 – hat{s}_a) M_a) is the number of reported cases at location a. Assuming that (p(t-1), p(t-2), ldots , p(0)) have been previously evaluated, the dynamics ((s_a, j_a)) is determined up to time t, and we have:
To obtain the scale p(t), we solve for:
We look for a simplified model in which the scales have a functional form close to a ramp function:
We first set (theta) by fitting this function to the daily scales:
Then we adjust the slope of the ramp in order to minimize the error with reported cases. In particular, we set (theta = (theta _1^*, theta _2^*, psi , theta _4^*)), and we solve:
We thus define (theta ^text {simplified} = (theta _1^*, theta _2^*, psi ^*, theta _4^*)), and the model for simplified scales is:
We also considered another approach, consisting of fitting the (N^2) transmission rates (beta _{ab}(t)) at every time t. In particular, we solve:
subject to the constraints (beta _{ab} ge 0). We show the results of this fit in Fig. S2, however this approach is prone to overfitting since there are (N^2) fitting parameters and only N data point at each time t.
The curves shown in Fig. 6b,c and the symbols shown in Fig. 6d were obtained by integrating Eq. (6).We used the function solve_ivp from SciPy with the “DOP853” integration method. At (t = 0), we considered (S = 1) everywhere except at the sites of coordinates ((0, 2^{m-1}-1)) and ((0, 2^{m-1})) (see Fig. 6a), where we set (S=0) and (I = 1). The Laplacian was computed using the 9-point stencil (Delta _text {discrete} = left( {begin{matrix} 1 &{} 2 &{} 1 \ 2 &{} -12 &{} 2 \ 1 &{} 2 &{} 1 end{matrix}}right) /4). We considered periodic boundary conditions along the y direction and Dirichlet boundary conditions along the x direction.
Data and scripts used in this study are available at the GitHub repository github.com/czbiohub/epidemiology_flux_model. Except the raw mobility data, which belongs to SafeGraph.
Henderson, J. Florence Under Siege: Surviving Plague in an Early Modern City (Yale University Press, 2019).
Google Scholar
Kermack, W. O. & McKendrick, A. G. A contribution to the mathematical theory of epidemics. Proc. R. Soc. Lond. Ser. A Contain. Pap. Math. Phys. Charact. 115, 700–721 (1927).
ADS MATH Google Scholar
Hethcote, H. W. The mathematics of infectious diseases. SIAM Rev. 42, 599–653. https://doi.org/10.1137/S0036144500371907 (2020).
ADS MathSciNet MATH Google Scholar
Harko, T., Lobo, F. S. & Mak, M. Exact analytical solutions of the susceptible-infected-recovered (SIR) epidemic model and of the SIR model with equal death and birth rates. Appl. Math. Comput. 236, 184–194 (2014).
MathSciNet MATH Google Scholar
Wu, J. T., Leung, K. & Leung, G. M. Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in Wuhan, China: A modelling study. Lancet 395, 689–697. https://doi.org/10.1016/S0140-6736(20)30260-9 (2020).
CAS PubMed PubMed Central Google Scholar
Kucharski, A. J. et al. Early dynamics of transmission and control of COVID-19: A mathematical modelling study. Lancet Infect. Dis. 20, 535–558. https://doi.org/10.1016/S1473-3099(20)30144-4 (2020).
Google Scholar
Allard, A., Moore, C., Scarpino, S. V., Althouse, B. M. & Hébert-Dufresne, L. The role of directionality, heterogeneity and correlations in epidemic risk and spread (2020). arXiv preprint arXiv:2005.11283.
Aleta, A. et al. Quantifying the importance and location of SARS-CoV-2 transmission events in large metropolitan areas. medRxiv https://doi.org/10.1101/2020.12.15.20248273 (2020).
PubMed PubMed Central Google Scholar
Hébert-Dufresne, L., Althouse, B. M., Scarpino, S. V. & Allard, A. Beyond (R_0): Heterogeneity in secondary infections and probabilistic epidemic forecasting. J. R. Soc. Interface 17, 20200393 (2020).
PubMed PubMed Central Google Scholar
Britton, T., Ball, F. & Trapman, P. A mathematical model reveals the influence of population heterogeneity on herd immunity to SARS-CoV-2. Science 369, 846–849 (2020).
ADS MathSciNet CAS PubMed PubMed Central MATH Google Scholar
Neipel, J., Bauermann, J., Bo, S., Harmon, T. & Jülicher, F. Power-law population heterogeneity governs epidemic waves. PLoS ONE 15, e0239678 (2020).
CAS PubMed PubMed Central Google Scholar
Sun, K. et al. Transmission heterogeneities, kinetics, and controllability of SARS-CoV-2. Science https://doi.org/10.1126/science.abe2424 (2021).
PubMed PubMed Central Google Scholar
Kawagoe, K. et al. Epidemic dynamics in inhomogeneous populations and the role of superspreaders. Phys. Rev. Res. 3, 033283. https://doi.org/10.1103/PhysRevResearch.3.033283 (2021).
CAS Google Scholar
Pozderac, C. & Skinner, B. Superspreading of SARS-CoV-2 in the USA. PLoS ONE 16, 1–10. https://doi.org/10.1371/journal.pone.0248808 (2021).
CAS Google Scholar
Huber, G. et al. A minimal model for household effects in epidemics. Phys. Biol. 17, 065010 (2020).
PubMed Google Scholar
Aleta, A. et al. Modelling the impact of testing, contact tracing and household quarantine on second waves of COVID-19. Nat. Hum. Behav. 4, 964–971. https://doi.org/10.1038/s41562-020-0931-9 (2020).
PubMed PubMed Central Google Scholar
Murray, J. Mathematical Biology II. Spatial Models and Biological Applications (Springer, 2003).
Google Scholar
Postnikov, E. B. & Sokolov, I. M. Continuum description of a contact infection spread in a SIR model. Math. Biosci. 208, 205–215 (2007).
MathSciNet PubMed MATH Google Scholar
Brockmann, D. & Helbing, D. The hidden geometry of complex, network-driven contagion phenomena. Science 342, 1337–1342 (2013).
ADS CAS PubMed Google Scholar
Chu, A., Huber, G., McGeever, A., Veytsman, B. & Yllanes, D. A random-walk-based epidemiological model. Sci. Rep. 11, 19308. https://doi.org/10.1038/s41598-021-98211-5 (2021).
ADS CAS PubMed PubMed Central Google Scholar
Te Vrugt, M., Bickmann, J. & Wittkowski, R. Effects of social distancing and isolation on epidemic spreading modeled via dynamical density functional theory. Nat. Commun. 11, 1–11 (2020).
Google Scholar
Tsori, Y. & Granek, R. Epidemiological model for the inhomogeneous spatial spreading of COVID-19 and other diseases. PLoS ONE 16, e0246056 (2021).
CAS PubMed PubMed Central Google Scholar
Tizzoni, M. et al. Real-time numerical forecast of global epidemic spreading: case study of 2009 A/H1N1pdm. BMC Med. 10, 1–31 (2012).
Google Scholar
Linka, K., Peirlinck, M., Sahli Costabal, F. & Kuhl, E. Outbreak dynamics of COVID-19 in Europe and the effect of travel restrictions. Comput. Methods Biomech. Biomed. Eng. 23, 710–717 (2020).
MATH Google Scholar
Ivorra, B., Ngom, D. & Ramos, A. M. Be-CoDiS: A mathematical model to predict the risk of human diseases spread between countries. validation and application to the 2014-15 Ebola virus disease epidemic. arXiv preprint arXiv:1410.6153 (2014).
Hsu, S. & Zee, A. Global spread of infectious diseases. J. Biol. Syst. 12, 289–300 (2004).
MATH Google Scholar
Nande, A., Adlam, B., Sheen, J., Levy, M. Z. & Hill, A. L. Dynamics of COVID-19 under social distancing measures are driven by transmission network structure. PLoS Comput. Biol. 17, e1008684 https://doi.org/10.1371/journal.pcbi.1008684 (2021).
ADS CAS PubMed PubMed Central Google Scholar
Ventura, P. C., Aleta, A., Rodrigues, F. A. & Moreno, Y. Modeling the effects of social distancing on the large-scale spreading of diseases. arXiv:2105.09697 (2021).
Mayberry, J., Nattestad, M. & Tuttle, A. The structure of an outbreak on a college campus. Math. Mag. 94, 83–98 (2021).
MathSciNet MATH Google Scholar
Chang, S. et al. Mobility network models of COVID-19 explain inequities and inform reopening. Nature 589, 82–87 (2021).
ADS CAS PubMed Google Scholar
Hethcote, H. W. An immunization model for a heterogeneous population. Theor. Popul. Biol. 14, 338–349 (1978).
MathSciNet CAS PubMed MATH Google Scholar
Post, W., DeAngelis, D. & Travis, C. Endemic disease in environments with spatially heterogeneous host populations. Math. Biosci. 63, 289–302 (1983).
MathSciNet MATH Google Scholar
May, R. M. & Anderson, R. M. Spatial heterogeneity and the design of immunization programs. Math. Biosci. 72, 83–111 (1984).
MathSciNet MATH Google Scholar
Hethcote, H. W. & Van Ark, J. W. Epidemiological models for heterogeneous populations: proportionate mixing, parameter estimation, and immunization programs. Math. Biosci. 84, 85–118 (1987).
MathSciNet MATH Google Scholar
Lloyd, A. L. & May, R. M. Spatial heterogeneity in epidemic models. J. Theor. Biol. 179, 1–11 (1996).
ADS CAS PubMed Google Scholar
SafeGraph. SafeGraph Social Distancing Metrics. https://docs.safegraph.com/docs/social-distancing-metrics (2021).
Dong, E., Du, H. & Gardner, L. An interactive web-based dashboard to track COVID-19 in real time. Lancet Infect. Dis. 20, 533–534 (2020).
CAS PubMed PubMed Central Google Scholar
Mehta, M. Random Matrices, 3rd ed, Sec. 1.5 (Elsevier, 2004).
Akemann, G., Baik, J. & Di Francesco, P. The Oxford Handbook of Random Matrix Theory (Oxford University Press, 2011).
MATH Google Scholar
Livan, G., Novaes, M. & Vivo, P. Introduction to Random Matrices: Theory and Practice (Springer, 2020).
MATH Google Scholar
Rosenzweig, N. & Porter, C. Repulsion of energy levels in complex atomic spectra. Phys. Rev. 120, 1698–1714 (1960).
ADS CAS Google Scholar
Girvan, M. & Newman, M. E. Community structure in social and biological networks. Proc. Natl. Acad. Sci. 99, 7821–7826 (2002).
ADS MathSciNet CAS PubMed PubMed Central MATH Google Scholar
Luo, F., Zhong, J., Yang, Y., Scheuermann, R. H. & Zhou, J. Application of random matrix theory to biological networks. Phys. Lett. A 357, 420–423 (2006).
ADS CAS Google Scholar
DeMeo, P., Ferrara, E., Fiumara, G. & Ricciardello, A. A novel measure of edge centrality in social networks. J. Knowl.-Based Syst. 30, 136–150 (2012).
Google Scholar
Brody, T. A. et al. Random-matrix physics: Spectrum and strength fluctuations. Rev. Mod. Phys. 53(385–479), 391 (1981).
ADS MathSciNet Google Scholar
Guhr, T., Müeller-Groeling, A. & Weidenmüller, H. A. Random matrix theories in quantum physics: Common concepts. Phys. Rep. 299(189–425), 228 (1998).
MathSciNet Google Scholar
Hakke, F. Quantum Signatures of Chaos 3rd edn, Sec 4.5 (Springer, 2010).
Naether, U., Postnikov, E. & Sokolov, I. Infection fronts in contact disease spread. Eur. Phys. J. B 65, 353–359 (2008).
ADS CAS Google Scholar
Dee, G. & Langer, J. Propagating pattern selection. Phys. Rev. Lett. 50, 383 (1983).
ADS Google Scholar
Ben-Jacob, E., Brand, H., Dee, G., Kramer, L. & Langer, J. Pattern propagation in nonlinear dissipative systems. Phys. D: Nonlinear Phenomena 14, 348–364 (1985).
ADS MathSciNet MATH Google Scholar
Bramson, M. Convergence of Solutions of the Kolmogorov Equation to Travelling Waves, Vol. 285 (American Mathematical Soc., 1983).
MATH Google Scholar
Paquette, G., Chen, L.-Y., Goldenfeld, N. & Oono, Y. Structural stability and renormalization group for propagating fronts. Phys. Rev. Lett. 72, 76 (1994).
ADS CAS PubMed Google Scholar
Ebert, U. & van Saarloos, W. Front propagation into unstable states: Universal algebraic convergence towards uniformly translating pulled fronts. Phys. D Nonlinear Phenomena 146, 1–99 (2000).
ADS MathSciNet MATH Google Scholar
Birzu, G., Hallatschek, O. & Korolev, K. S. Fluctuations uncover a distinct class of traveling waves. Proc. Natl. Acad. Sci. 115, E3645–E3654 (2018).
ADS MathSciNet CAS PubMed PubMed Central MATH Google Scholar
Sen, P., Yamana, T. K., Kandula, S., Galanti, M. & Shaman, J. Burden and characteristics of COVID-19 in the united states during 2020. Nature 598, 338–341 (2021).
ADS CAS Google Scholar
SafeGraph. SafeGraph Neighborhood Patterns. https://docs.safegraph.com/docs/neighborhood-patterns (2021).
SafeGraph. SafeGraph Open Census Data. https://docs.safegraph.com/docs/open-census-data (2021).
Download references
The authors acknowledge the generous support of the Chan Zuckerberg Initiative and Chan Zuckerberg Biohub. The authors also thank Mark Rychnovsky for useful conversations on random matrix theory in the early stages of this project, and Naomi Rankin for her help with a preliminary version of the lattice simulations. D.Y. acknowledges support by MINECO (Spain) through Grant No. PGC2018-094684-B-C21, partially funded by the European Regional Development Fund (FEDER). We also are grateful to the anonymous reviewers of Scientific reports who asked deep questions, that helped us to better understand our results.
Chan Zuckerberg Biohub, 499 Illinois Street, San Francisco, CA, 94158, USA
Guillaume Le Treut, Greg Huber, Mason Kamb, Aaron McGeever & David Yllanes
Department of Physics, Kadanoff Center for Theoretical Physics, University of Chicago, Chicago, IL, 60637, USA
Kyle Kawagoe
Okinawa Institute of Science and Technology, Onna-son, Okinawa, 904-0495, Japan
Jonathan Miller & Reuven Pnini
Chan Zuckerberg Initiative, Redwood City, CA, 94063, USA
Boris Veytsman
School of Systems Biology, George Mason University, Fairfax, VA, 22030, USA
Boris Veytsman
Instituto de Biocomputación y Física de Sistemas Complejos (BIFI), 50018, Zaragoza, Spain
David Yllanes
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
The concept for this research grew out of discussions among all authors; G.L.T. performed research; G.H., A.M., M.K. analyzed data; K.K., J.M., R.P., performed analytical/numerical computations; G.L.T., B.V. and D.Y. wrote the paper, with contributions from all authors.
Correspondence to Guillaume Le Treut or Greg Huber.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Video S1.
Supplementary Video S2.
Supplementary Video S3.
Supplementary Video S4.
Supplementary Video S5.
Supplementary Video S6.
Supplementary Video S7.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Le Treut, G., Huber, G., Kamb, M. et al. A high-resolution flux-matrix model describes the spread of diseases in a spatial network and the effect of mitigation strategies. Sci Rep 12, 15946 (2022). https://doi.org/10.1038/s41598-022-19931-w
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-19931-w
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
Download Point Blank’s Free Vocoder Plugin – Watch Tutorial Now – Magnetic Magazine
Homer Dudley initially developed vocoders at the Bell Labs in the 1930s, whilst he was experimenting with speech synthesis. Vocoders work using a modulator signal, like a voice or drum beat, and a carrier signal, such as a sustained pad or synth sound.
These signals are then split through a number of bandpass filters that combine and output the common frequency between the two signals. An iconic use of that robotic-vocal sound, created by a vocoder, can be found in Kraftwerk’s 1978 track, ‘The Robots‘.
If you’re keen to learn more music production skills, much like this one, then be sure to check out Point Blank’s music production courses in Los Angeles. Alternatively, if you’re looking to study with Point Blank in London or online, they have a range of degree courses that specialize in music production and sound engineering. Check out their full roster here.
During the video, Dan demonstrates how PB’s new vocoder works, showing how you can transform a drum beat into a robotic vocal sound. He runs through the different parameters, revealing how you can get the most out of Point Blank’s free vocoder plugin.
The new Max for Live plugin by Dan Herbert is available exclusively from Point Blank – it will only cost you your email address. To find help on how to install the new Max for Live vocoder plugin, be sure to watch their previous Kick Drum Designer video.
Researchers Uncover Years-Long Mobile Spyware Campaign Targeting Uyghurs – The Hacker News
A new wave of a mobile surveillance campaign has been observed targeting the Uyghur community as part of a long-standing spyware operation active since at least 2015, cybersecurity researchers disclosed Thursday.
The intrusions, originally attributed to a threat actor named Scarlet Mimic back in January 2016, is said to have encompassed 20 different variants of the Android malware, which were disguised as books, pictures, and an audio version of the Quran.
The malware, while relatively unsophisticated from a technical standpoint, comes with extensive capabilities to steal sensitive data from an infected device, send SMS messages on the victim’s behalf, make phone calls, and track their locations.
Additionally, it allows the recording of incoming and outgoing phone calls as well as surrounding audio.
“All this makes it a powerful and dangerous surveillance tool,” Israeli cybersecurity firm Check Point said in a technical deepdive, calling the spyware MobileOrder.
It’s worth noting that a part of the campaign was recently disclosed by researchers from the MalwareHunterTeam and Cyble, in which a book written by the exiled Uyghur leader Dolkun Isa was used as a lure to deliver the malware.
Check Point said it observed MobileOrder artifacts in the wild right from 2015 to mid-August 2022, with the exception of 2021, when none were detected.
Attack campaigns likely involve the use of social engineering tactics to trick unsuspecting victims into launching malicious applications that reference seemingly innocuous documents, photos, and audio files.
These apps contain a variety of baits, including a PDF about guerrilla warfare and pictures related to the deployment of paramilitary forces in Ürümqi, the capital of the Xinjiang Uyghur Autonomous Region, in the aftermath of the deadly April 2014 attack.
Opening the rogue app, in turn, launches a decoy document designed to distract the target from noticing the malicious actions in the background.
“Some of the versions also ask for Device Admin and root access, which not only gives the malware full access to the device, but also prevents the victim from easily uninstalling the application,” the researchers said.
Other features supported by MobileOrder include executing a remote shell and even dropping additional Android Package (APK) files.
The campaign’s attribution to Scarlet Mimic, per Check Point, stems from clear code overlaps, shared infrastructure, and the same victimology patterns.
Furthermore, the ongoing use of MobileOrder signals a shift in attack vector from desktop to mobile surveillance, what with the actor previously linked to a Windows malware called Psylo Trojan.
While it’s not clear which of these attacks throughout the past seven years have been successful, the very fact that the malware authors are continuing to deploy the spyware is an indication that some of these efforts have paid off.
“The persistence of the campaign, the evolution of the malware, and the persistent focus on targeting specific populations indicate that the group’s operations over the years are successful to some extent,” Check Point said.
Sign up for cybersecurity newsletter and get latest news updates delivered straight to your inbox daily.

