Kaustubha Mendhurwar

Edge-Detection in Noisy Images Using Independent Component Analysis

Edges in a digital image provide important information about the objects contained within the image since they constitute boundaries between objects in the image. This paper proposes a new approach based on independent component analysis (ICA) for edge-detection in noisy images. The proposed approach works in two phases—the training phase and the edge-detection phase. The training phase is carried out only once to determine parameters for the ICA. Once calculated, these ICA parameters can be employed for edge-detection in any number of noisy images. The edge-detection phase deals with transitioning in and out of ICA domain and recovering the original image from a noisy image. Both gray scale as well as colored images corrupted with Gaussian noise are studied using the proposed approach, and remarkably improved results, compared to the existing edge-detection techniques, are achieved. Performance evaluation of the proposed approach using both subjective as well as objective methods is presented.

Microarray Image Denoising using Independent Component Analysis

DNA microarrays have proved to be one of the vital breakthrough technologies for exploring the patterns of gene expression on a global scale. The differential measured gene-expression levels depend largely on the probe intensities extracted during microarray image processing. Various noises introduced during the experiment and the imaging process can drastically influence the accuracy of results. Microarray image denoising is one of the challenging preprocessing steps in microarray image analysis. In this paper, we propose denoising of microarray images using the independent component analysis (ICA). The idea of ICA i.e. finding the linear representation of nongaussian data so that the components are independent (or atleast as independent as possible) is exploited for denoising microarray images. Through examples, it is shown that the proposed approach is highly effective as compared to the conventional discrete wavelet transform and statistical methods.

Microarray image segmentation using chan-vese active contour model and level set method

Microarray technology is considered to be one of the major breakthroughs in bioinformatics for profiling gene-expressions of thousands of genes, simultaneously. Analysis of a microarray image plays an important role in the accurate depiction of gene-expression. Segmentation, the process of separating the foreground from the background, of a microarray image, is one of the key issues in microarray image analysis. Level sets have tremendous potential in the segmentation of images. In this paper, a new approach for segmentation of the microarray images is proposed. In this work, Chan-Vese approximation of the Mumford-Shah model and the level set method are employed for image segmentation. Illustrative examples of the proposed method are presented highlighting its effectiveness.

Binning algorithm for accurate computer aided device modeling

Accurate modeling of devices is critical to efficient computer aided design and optimization. Commonly encountered modeling techniques include empirical formulae, equivalent circuits, and black-box models (eg. neural networks). Important criteria in device modeling are model accuracy, computational simplicity, generality of the modeling approach, and so forth. In this paper, we present a new and systematic CAD algorithm to device modeling based on a concept often referred to as binning. For a given set of data either from measurements or simulations, the proposed algorithm leads to an accurate model comprising of a set of sub-models with best possible accuracy, while keeping the model structure simple. The proposed algorithm is general and can be applied in the context of any black-box modeling technique. In this paper, the algorithm is illustrated for the case of neural network modeling. Resulting models are shown to exhibit relatively better accuracies compared to those developed using a standard modeling approach. Both active and passive modeling examples are presented.

A new neural network modeling approach based on a correction model concept

Neural networks have recently gained attention as unconventional yet effective alternatives for component modeling. One of the most commonly used neural networks, namely the multilayer perceptrons (MLP) could sometimes fail to model highly nonlinear input-output behaviors accurately. Advanced neural networks (e.g. knowledge based neural networks) can be employed; however, such networks suffer from an increased complexity both in terms of their structures and training methods. In this paper, we propose a neural network modeling approach based on a novel correction model concept. This approach helps accurately model complicated behaviors using simple 3-layer MLP networks. Both active and passive examples are presented.

A new computer-aided multi-dimensional device modeling algorithm based on binning concepts

Accurate modeling of devices is critical to efficient computer aided design and optimization. Commonly encountered modeling techniques include empirical formulae, equivalent circuits, and black-box models. Important criteria in device modeling are model accuracy, computational simplicity, generality of the modeling approach, etc. In this paper, we present a multi-dimensional CAD algorithm to device modeling based on a concept often referred to as binning. For a given set of data either from measurements or simulations, the proposed algorithm leads to an accurate model comprising of a set of sub-models with best possible accuracy, while keeping the model structure simple. The algorithm is general and can be applied in the context of any black-box modeling technique; however, we demonstrate multi-dimensional neural modeling of active and passive components. Resulting models are shown to exhibit relatively better accuracies compared to those developed via standard modeling approaches.

Learning human action sequence style from video for transfer to 3D game characters

In this paper, we present an innovative framework for a 3D game character to adopt human action sequence style by learning from videos. The framework is demonstrated for kickboxing, and can be applied to other activities in which individual style includes improvisation of the sequence in which a set of basic actions are performed. A video database of a number of actors performing the basic kickboxing actions is used for feature word vocabulary creation using 3D SIFT descriptors computed for salient points on the silhouette. Next an SVM classifier is trained to recognize actions at frame level. Then an individual actors action sequence is gathered automatically from the actors kickboxing videos and an HMM structure is trained. The HMM, equipped with the basic repertoire of 3D actions created just once, drives the action level behavior of a 3D game character.