Ninad Thakoor

Hidden Markov Model-Based Weighted Likelihood Discriminant for 2-D Shape Classification

The goal of this paper is to present a weighted likelihood discriminant for minimum error shape classification. Different from traditional maximum likelihood (ML) methods, in which classification is based on probabilities from independent individual class models as is the case for general hidden Markov model (HMM) methods, proposed method utilizes information from all classes to minimize classification error. The proposed approach uses a HMM for shape curvature as its 2-D shape descriptor. We introduce a weighted likelihood discriminant function and present a minimum classification error strategy based on generalized probabilistic descent method. We show comparative results obtained with our approach and classic ML classification with various HMM topologies alongside Fourier descriptor and Zernike moments-based support vector machine classification for a variety of shapes.

Vision and Attention Theory Based Sampling for Continuous Facial Emotion Recognition

Affective computing-the emergent field in which computers detect emotions and project appropriate expressions of their own-has reached a bottleneck where algorithms are not able to infer a persons emotions from natural and spontaneous facial expressions captured in video. While the field of emotion recognition has seen many advances in the past decade, a facial emotion recognition approach has not yet been revealed which performs well in unconstrained settings. In this paper, we propose a principled method which addresses the temporal dynamics of facial emotions and expressions in video with a sampling approach inspired from human perceptual psychology. We test the efficacy of the method on the Audio/Visual Emotion Challenge 2011 and 2012, CohnKanade and the MMI Facial Expression Database. The method shows an average improvement of 9.8 percent over the baseline for weighted accuracy on the Audio/Visual Emotion Challenge 2011 video-based frame-level subchallenge testing set.

Branch-and-Bound for Model Selection and Its Computational Complexity

Branch-and-bound methods are used in various data analysis problems, such as clustering, seriation and feature selection. Classical approaches of branch-and-bound based clustering search through combinations of various partitioning possibilities to optimize a clustering cost. However, these approaches are not practically useful for clustering of image data where the size of data is large. Additionally, the number of clusters is unknown in most of the image data analysis problems. By taking advantage of the spatial coherency of clusters, we formulate an innovative branch-and-bound approach, which solves clustering problem as a model-selection problem. In this generalized approach, cluster parameter candidates are first generated by spatially coherent sampling. A branch-and-bound search is carried out through the candidates to select an optimal subset. This paper formulates this approach and investigates its average computational complexity. Improved clustering quality and robustness to outliers compared to conventional iterative approach are demonstrated with experiments.

A statistical approach for intensity loss compensation of confocal microscopy images

In this paper, a probabilistic technique for compensation of intensity loss in confocal microscopy images is presented. For single‐colour‐labelled specimen, confocal microscopy images are modelled as a mixture of two Gaussian probability distribution functions, one representing the background and another corresponding to the foreground. Images are segmented into foreground and background by applying Expectation Maximization algorithm to the mixture. Final intensity compensation is carried out by scaling and shifting the original intensities with the help of parameters estimated for the foreground. Since foreground is separated to calculate the compensation parameters, the method is effective even when image structure changes from frame to frame. As intensity decay function is not used, complexity associated with estimation of the intensity decay function parameters is eliminated. In addition, images can be compensated out of order, as only information from the reference image is required for the compensation of any image. These properties make our method an ideal tool for intensity compensation of confocal microscopy images that suffer intensity loss due to absorption/scattering of light as well as photobleaching and the image can change structure from optical/temporal section‐to‐section due to changes in the depth of specimen or due to a live specimen. The proposed method was tested with a number of confocal microscopy image stacks and results are presented to demonstrate the effectiveness of the method.

Branch-and-bound hypothesis selection for two-view multiple structure and motion segmentation

An efficient and robust framework for two-view multiple structure and motion segmentation is proposed. To handle this otherwise recursive problem, hypotheses for the models are generated by local sampling. Once these hypotheses are available, a model selection problem is formulated which takes into account the hypotheses likelihoods and model complexity. An explicit model for outliers is also added for robust model selection. The model selection criterion is optimized through branch-and-bound technique of combinatorial optimization which guaranties optimality over current set of hypotheses by efficient search of solution space.

Shape classifier based on generalized probabilistic descent method with hidden Markov descriptor

The goal of this paper is to present a weighted likelihood discriminant for minimum error shape classification. Different from traditional maximum likelihood (ML) methods, in which classification is based on probabilities from independent individual class models as is the case for general hidden Markov model (HMM) methods, proposed method utilizes information from all classes to minimize classification error. The proposed approach uses a HMM for shape curvature as its 2D shape descriptor. In this contribution, we introduce a weighted likelihood discriminant function and present a minimum error classification strategy based on generalized probabilistic descent (GPD) method. We believe our sound theory based implementation reduces classification error by combining HMM with GPD theory. We show comparative results obtained with our approach and classic ML classification along with Fourier descriptor and Zernike moments based classification for fighter planes and vehicle shapes.

Real-time Planar Surface Segmentation in Disparity Space

An iterative Segmentation-Estimation framework for segmentation of planar surfaces in the disparity space is implemented on a Digital Signal Processor (DSP). Disparity of a scene is modeled by approximating various surfaces in the scene to be planar. The surface labels are estimated during the segmentation phase of the framework with help of the underlying plane parameters. After segmentation, planar surfaces are separated into spatially continuous regions. The largest of these regions is used to compute the estimates for the plane parameters. The iterative process is continued till convergence. The algorithm was optimized and implemented on TMS320DM642 based embedded system that operates at 3 to 5 frames per second on images of size 320 x 240.

Structural Signatures for Passenger Vehicle Classification in Video

This paper focuses on a challenging pattern recognition problem of significant industrial impact, i.e., classifying vehicles from their rear videos as observed by a camera mounted on top of a highway with vehicles traveling at high speed. To solve this problem, this paper presents a novel feature called structural signature. From a rear-view video, a structural signature recovers the vehicle side profile information, which is crucial in its classification. As a vehicle moves away from a camera, its surfaces deform differently based on their relative orientation to the camera. This information is used to extract the structure of a vehicle, which captures the relative orientation of vehicle surfaces and the road surface. This paper presents a complete system that computes structural signatures and uses them for classification of passenger vehicles into sedans, pickups, and minivans/sport utility vehicles in highway videos. It analyzes the performance of the proposed system on a large video data set.

Automatic video object shape extraction and its classification with camera in motion

In this paper, we present an automatic moving object extraction and classification system. For automatic extraction of object taken by a moving camera, a novel technique is proposed, in which optical flow handles the background modeling and camera motion estimation, and frame difference information yields the exact object shape. We also use forward region boundary based change detection approach for frame difference. This approach assures change detection for uniform intensity regions. For classification, weighted likelihood discriminant based shape classifier is designed. Unlike maximum likelihood (ML) methods, our proposed method utilizes information from all classes to design the classifier. In the description phase of the classifier, curvature features are extracted from the shape and are utilized to build a hidden Markov model (HMM). The HMM provides a robust ML description of the shape. In the discrimination phase, a weighted likelihood discriminant function is introduced, which weights the likelihoods of curvature at individual points of the shape to minimize the classification error. The weights are estimated by generalized probabilistic descent (GPD) method. To demonstrate the performance of the proposed method, we present results achieved for car shapes extraction and classification.

Multibody Structure-and-Motion Segmentation by Branch-and-Bound Model Selection

An efficient and robust framework is proposed for two-view multiple structure-and-motion segmentation of unknown number of rigid objects. The segmentation problem has three unknowns, namely the object memberships, the corresponding fundamental matrices, and the number of objects. To handle this otherwise recursive problem, hypotheses for fundamental matrices are generated through local sampling. Once the hypotheses are available, a combinatorial selection problem is formulated to optimize a model selection cost which takes into account the hypotheses likelihoods and the model complexity. An explicit model for outliers is also added for robust segmentation. The model selection cost is minimized through the branch-and-bound technique of combinatorial optimization. The proposed branch-and-bound approach efficiently searches the solution space and guaranties optimality over the current set of hypotheses. The efficiency and the guarantee of optimality of the method is due to its ability to reject solutions without explicitly evaluating them. The proposed approach was validated with synthetic data, and segmentation results are presented for real images.