This paper presents a generative statistical model for analyzing time series of planar shapes. Using elastic shape analysis, we separate object kinematics (rigid motions and speed variability) from morphological evolution, representing the latter through transported velocity fields (TVFs). A principal component analysis (PCA) based dimensionality reduction of the TVF representation provides a finite-dimensional Euclidean framework, enabling traditional time-series analysis. We then fit a vector auto-regressive (VAR) model to the TVF-PCA time series, capturing the statistical dynamics of shape evolution. To characterize morphological changes, we use VAR model parameters for model comparison, synthesis, and sequence classification. Leveraging these parameters, along with machine learning classifiers, we achieve high classification accuracy. Extensive experiments on cell motility data validate our approach, demonstrating its effectiveness in modeling and classifying migrating cells based on morphological evolution—marking a novel contribution to the field. • A novel method for statistical modeling and analysis of temporally-evolving shapes. • Elastic shape analysis represents a shape sequence by Euclidean time series. • Dynamic shape evolution is crucial aspect of cell migration in biology. • A detailed approach for modeling and classifying shape dynamics during cell motility.

The method proposed in this paper is a robust combination of multi-task learning and unsupervised domain adaptation for segmenting amoeboid cells in microscopy. A highlight of this work is the manner in which the model's hyperparameters are estimated. The detriments of ad-hoc parameter estimation are well known, but this issue remains largely unaddressed in the context of CNN-based segmentation. Using a novel min-max formulation of the segmentation cost function our proposed method analytically estimates the model's hyperparameters, while simultaneously learning the CNN weights during training. This end-to-end framework provides a consolidated mechanism to harness the potential of multi-task learning to isolate and segment clustered cells from low contrast brightfield images, and it simultaneously leverages deep domain adaptation to segment fluorescent cells without explicit pixel-level re- annotation of the data. Experimental validations on multi-cellular images strongly suggest the effectiveness of the proposed technique, and our quantitative results show at least 15% and 10% improvement in cell segmentation on brightfield and fluorescence images respectively compared to contemporary supervised segmentation methods.

This paper addresses the problem of characterizing statistical distributions of cellular shape populations using shape samples from microscopy image data. This problem is challenging because of the nonlinearity and high-dimensionality of shape manifolds. The paper develops an efficient, nonparametric approach using ideas from k-modal mixtures and kernel estimators. It uses elastic shape analysis of cell boundaries to estimate statistical modes and clusters given shapes around those modes. (Notably, it uses a combination of modal distributions and ANOVA to determine k automatically.) A population is then characterized as k-modal mixture relative to this estimated clustering and a chosen kernel (e.g., a Gaussian or a flat kernel). One can compare and analyze populations using the Fisher-Rao metric between their estimated distributions. We demonstrate this approach for classifying shapes associated with migrations of entamoeba histolytica under different experimental conditions. This framework remarkably captures salient shape patterns and separates shape data for different experimental settings, even when it is difficult to discern class differences visually.

Statistical analysis of shape evolution during cell migration is important for gaining insights into biological processes. This paper develops a time-series model for temporal evolution of cellular shapes during cell motility. It uses elastic shape analysis to represent and analyze shapes of cell boundaries (as planar closed curves), thus separating cell shape changes from cell kinematics. Specifically, it utilizes Transported Square-Root Velocity Field (TSRVF), to map non-Euclidean shape sequences into a Euclidean time series. It then uses PCA to reduce Euclidean dimensions and imposes a Vector Auto-Regression (VAR) model on the resulting low-dimensional time series. Finally, it presents some results from VAR-based statistical analysis: estimation of model parameters and diagnostics, synthesis of new shape sequences, and predictions of future shapes given past shapes.

Detecting and segmenting individual cells from microscopy images is critical to various life science applications. Traditional cell segmentation tools are often ill-suited for applications in brightfield microscopy due to poor contrast and intensity heterogeneity, and only a small subset are applicable to segment cells in a cluster. In this regard, we introduce a novel supervised technique for cell segmentation in a multitask learning paradigm. A combination of a multi-task loss, based on the region and cell boundary detection, is employed for an improved prediction efficiency of the network. The learning problem is posed in a novel min-max framework which enables adaptive estimation of the hyper-parameters in an automatic fashion. The region and cell boundary predictions are combined via morphological operations and active contour model to segment individual cells. The proposed methodology is particularly suited to segment touching cells from brightfield microscopy images without manual interventions. Quantitatively, we observe an overall Dice score of 0.93 on the validation set, which is an improvement of over 15.9% on a recent unsupervised method, and outperforms the popular supervised U-net algorithm by at least 5.8% on average.

Recent deep learning based approaches have outperformed classical stereo matching methods. However, current deep learning based end-to-end stereo matching methods adopt a generic encoder-decoder style network with skip connections. To limit computational requirement, many networks perform excessive down sampling, which results in significant loss of useful low-level information. Additionally, many network designs do not exploit the rich multi-scale contextual information. In this work, we address these aforementioned problems by carefully designing the network architecture to preserve required spatial information throughout the network, while at the same time achieve large effective receptive field to extract multiscale contextual information. For the first time, we create a synthetic disparity dataset reflecting real life images captured using a smartphone; this enables us to obtain state-of-the-art results on common real life images. The proposed model DISCO is pre-trained on the synthetic Scene Flow dataset and evaluated on popular benchmarks and our in-house dataset of challenging real life images. The proposed model outperforms existing state-of-the-art methods in terms of quality as well as quantitative metrics.

The rapid development of camera in hand-held devices and the emergence of social media has led to an uprise in capturing self-portrait images. Augmenting these images for beautification or applying special effects to mimic DSLR camera has become a popular practice. Most of these effects require separation of foreground from background where the effect can be applied solely on background. To employ such effects on portrait (upper half of human body) images, a pixel-accurate segmentation is imperative. In this paper, we propose an effective method of fast depth aware CNN based portrait segmentation from monocular images. The proposed method is capable of being deployed on mobile phones, within the constraints of time and memory. On the segmented images, we demonstrate the application of bokeh effect, which blurs out-of-focus regions. We experiment with different combinations of state of the art encoder and decoder networks for segmentation and infer that our proposed method can improve the inference speed by 76 ms on mobile device while maintaining an accuracy of 97.0 %.

In image classification, obtaining adequate data to learn a robust classifier has often proven to be difficult in several scenarios. Classification of histological tissue images for health care analysis is a notable application in this context due to the necessity of surgery, biopsy or autopsy. To adequately exploit limited training data in classification, we propose a saliency guided dictionary learning method and subsequently an image similarity technique for histo-pathological image classification. Salient object detection from images aids in the identification of discriminative image features. We leverage the saliency values for the local image regions to learn a dictionary and respective sparse codes for an image, such that the more salient features are reconstructed with smaller error. The dictionary learned from an image gives a compact representation of the image itself and is capable of representing images with similar content, with comparable sparse codes. We employ this idea to design a similarity measure between a pair of images, where local image features of one image, are encoded with the dictionary learned from the other and vice versa. To effectively utilize the learned dictionary, we take into account the contribution of each dictionary atom in the sparse codes to generate a global image representation for image comparison. The efficacy of the proposed method was evaluated using three tissue data sets that consist of mammalian kidney, lung and spleen tissue, breast cancer, and colon cancer tissue images. From the experiments, we observe that our methods outperform the state of the art with an increase of 14.2% in the average classification accuracy over all data sets.

Short-time event detection from videos obtained using handheld or car-mounted cameras is an overarching challenge in surveillance. The problem demands simultaneous spatiotemporal localization of the event along with removal of a dynamic background. Existing state-of-the-art techniques are sensitive to non-uniform jitter, changing background, and clutter. In this paper, we propose graph Laplacian assisted parametric dictionary learning, GraDED to account for the aforementioned variations. The temporal occurrence and duration of the event is determined from weights learned using a dynamic graph, while the spatial localization is performed by graph based dictionary learning. We demonstrate the efficacy of our approach by comparing with three state-of-the-art methods and achieve on average an overall increase of 0.08 in specificity and 0.6 in sensitivity for event detection.

In microscopy, new super-resolution methods are emerging that produce three-dimensional images at resolutions ten times smaller than that provided by traditional light microscopy. Such technology is enabling the exploration of structure and function in living tissues such as bacterial biofilms that have mysterious interconnections and organization. Unfortunately, the standard tools used in the image analysis community to perform segmentation and other higher-level analyses cannot be applied naïvely to these data. This paper presents Bact-3D, a 3D method for segmenting super-resolution images of multi-leveled, living bacteria cultured in vitro. The method incorporates a novel initialization approach that exploits the geometry of the bacterial cells as well an iterative local level set evolution that is tailored to the biological application. In experiments where segmentation is used as precursor to cell detection, the Bact-3D matches or improves upon the Dice score and mean-squared error of two existing methods, while yielding a substantial improvement in cell detection accuracy. In addition to providing improvements in performance over the state-of-the-art, this report also characterizes the tradeoff between imaging resolution and segmentation quality.

In this paper we present a novel idea of evaluating similarity between two images aided by a salient object detection framework. For computing similarity between images consisting of multiple objects and varying background, extracting features relevant to the object of interest is of cruicial importance. To accomplish this task, we employ a saliency guided dictionary learning framework for image similarity evaluation (SLIDE). The saliency detection framework emphasizes the image regions that attract human attention and is exploited to build a dictionary for generating sparse representation of the images. The compressibility of the sparse codes is exploited in computing the similarity measure. The SLIDE framework is used in the application of image retrieval for three military image datasets and shows in average an improvement of 14.8% in retrieval accuracy compared to two state-of-the-art similarity measures - information distance and sparse representation based compression distance.

We propose a novel region based segmentation technique using dictionary learning. In a previous work we have developed a method which uses a set of pre-specified Legendre basis functions to perform region based segmentation of an object in presence of heterogeneous illumination. We hypothesize that in problems where a set of training images for the object is available for analysis (such as depth image sequence of blood vessels via ultrasound imaging), segmentation accuracy can be significantly improved by learning the basis functions instead of specifying them implicitly. The salient idea of this letter is to compute the optimal set of functions to model the region intensities. Our solution to this problem involves the integration of a level set segmentation methodology with the dictionary learning framework. This provides an elegant solution to deal with intensity inhomogeneities prevalent in many imaging applications such as ultrasound and fluorescence microscopy. The proposed algorithm, Dictionary Learning Level Set (DL2S) is used to segment ultrasound images of blood vessels captured using low cost, portable ultrasound devices employed in a phlebotomy application. Qualitative and quantitative results obtained from this dataset suggest efficacy of D2LS with an associated improvement in the average Dice index of 12% over the relevant competitors.

With increasing complexity of the dataset it becomes impractical to use a single feature to characterize all constituent images. In this paper we describe a method that will automatically select the appropriate image features that are relevant and efficacious for classification, without requiring modifications to the feature extracting methods or the classification algorithm. We first describe a method for designing class distinctive dictionaries using a dictionary learning technique, which yields class specific sparse codes and a linear classifier parameter. Then, we apply information theoretic measures to obtain the more informative feature relevant to a test image and use only that feature to obtain final classification results. With at least one of the features classifying the query accurately, our algorithm chooses the correct feature in 88.9% of the trials.

In this work, we develop algorithms for tracking time sequences of sparse spatial signals with slowly changing sparsity patterns, and other unknown states, from a sequence of nonlinear observations corrupted by (possibly) non-Gaussian noise. A key example of the above problem occurs in tracking moving objects across spatially varying illumination changes, where motion is the small dimensional state while the illumination image is the sparse spatial signal satisfying the slow-sparsity-pattern-change property.