Search

The article discusses a new approach, where in the Gauss-Hermite (GH) expression is incorporated in the probability density function (PDF) for the denoising of the Bayesian wavelet-based image. The new PDF is shown to be extremely effective in the subband-adaptive, as well as the locally adaptive conditions.

The article explains a newly developed Bayesian algorithm, which uses a previously defined image formed by imposing the Student t-densities on the outputs for the image restoration problem. The approach is shown to be extremely advantageous, as it does not require the empire parameter selection.

An automatic classification framework consisting of appearance based features and hidden Markov models (HMM) is proposed for the detection of unusual events in image sequences. Findings reveal a reduction in the feature space dimension of the image sequences.

Sylvester-Lyapunov equation is a well-known mathematical framework for the unification of a set of Bayesian denoising algorithms for additive and multiplicative noise. Findings reveal the achievement of competitive results with the best tested methods in a wide variety of experimental conditions.

An extended version of Stein's unbiased risk estimate (SURE) is proposed for performing optimization adaptively for each observed noisy image. Findings reveal the better performance of SURE in case of overcomplete subbands.

A method for the detection of humans in a single image based on a novel cascaded structure is discussed. Results suggest the achievement of high detection accuracy and efficiency.

A probabilistic framework for tracking multiview faces and learning their appearance model online is presented. Results suggest the capability of the proposed framework in robustly tracking oblique pose changes and simultaneously estimating pose.

A new simple closed-form parametric multivariate prior framework that will match well with the observed statistics of the wavelet coefficients of natural images is developed. The empirical outcomes demonstrate the effectiveness of the model in the wavelet-based Bayesian image estimation and enhanced estimation performance over related techniques.

The article presents a strategy for image segmentation by active contours. Results suggest the versatility and flexibility of the strategy for accommodating a diversity of image features for segmentation.

Affine subspace for verification (ASV) is a novel algorithm for verifying the subspace learning techniques. Findings reveal the effectiveness of ASV in achieving encouraging face verification accuracy.

A Bayesian wavelet-based denoising procedure for multicomponent images which makes use of Gaussian scale mixtures as prior models that approximate the marginal distribution of the wavelet coefficients is presented.

A new technique is proposed to compute the solution efficiently for a common class of nonlinear algorithms for edge-preserving image restoration. The algorithm exploits the sparsity of edges to define an FFT (fast Fourier transforms)-based iteration that requires few iterations and is guaranteed to converge to the MAP (maximum a posteriori) estimate.

The simultaneous autoregressions are used as prior distributions for both the image and blur, and gamma distributions for the unknown parameters of the prions and the image formation noise, following the hierarchical Bayesian framework for blind deconvolution problems. The proposed algorithms are tested and compared with the existing blind deconvolution methods.

A fully automated procedure for automatic microarray gridding, based on the Bayesian paradigm and image analysis techniques, is presented that is aimed at generating a grid hypothesis based on the Random transform, and which accounts for local grid deformations. The method could deal with various kinds of perturbation such as image rotations, spot irregularities and deviations, and experiments over synthetic and real images suggest that it outperforms other methods for noisy images.

Two image restoration algorithms based on the Bayesian approach and a new hierarchical spatially adaptive image prior are proposed. Numerical experiments are conducted to compare the algorithms with other stationary and nonstationary algorithms, which demonstrate the efficiency of the proposed algorithms.

A Hidden Markov Random Field (HMRF) data-fusion model based on Julesz ensembles is presented and applied to the segmentation of natural images. The corresponding Exploration/Selection/Estimation (ESE) procedure for the estimation of the parameters is presented.

Explicit lip motion information is proposed to be used, instead of or in addition to lip intensity and/or geometry information, for speaker identification and speech-reading. A novel two-stage, spatial and temporal discrimination analysis framework is introduced to select the best lip motion features, and experiments using an hidden-Markov-model-based recognition system suggest that explicit lip motion information provides performance gains in both the applications, with lip motion features more useful in the speech-reading application.

The blind source separation (BSS) of images from noisy linear mixtures with unknown coefficient formulated as a Bayesian estimation problem is discussed. It is proved that the source models accounting for local correlation could increase the robustness against noise and also against cross correlation of the sources and the derived formula is useful for the segmentation and classification of images.

The problem of simultaneous structure from motion estimation for multiple independently moving objects from a monocular image sequence is addressed. Two Bayesian algorithms are presented for solving the problem that uses the sequential importance sampling (SIS) technique.

A 3-D contextual classification algorithm, based on Owen, Hjort amd Mohn's 2-d contextual classification algorithm, is described.

Research describing a method for estimating multivariate mixtures is presented. The proposed extension of the method of generalized mixture estimation has implications for signal processing applications.

The prototype Bayesian image retrieval system, PicHunter, is discussed relative to implementation, theory and psychophysical experiments. PicHunter is a content-based image retrieval (CBIR) system recently developed. With an explicit model of what users would do, given the target image, the system uses Bayes's rule to predict the target, given the user's actions. An entropy-minimizing display algorithm that tries to maximize the information for a user at each iteration of the search is given. Hidden annotation vs a learned system is used.

The first steps toward the long-term goal of identifying semantically relevant features and building computational models for extraction and characterization would be developing statistical models for two important elements of the video structure, shot duration and activity. These are are discussed with the argument that knowledge about structure can be used to improve performance of content analysis and to extract features that convey semantic content information. A Bayesian formulation for the shot segmentation problem has been developed. It extends the standard thresholding model in an adaptive/ intuitive way and leads to better segmentation. The Bayesian model captures semantic properties of content. Ways to use these properties for intuitive content-based access to movie libraries exist.

A hypothesis test based on the Bayes decision criterion was developed in order to identify stationary, moving and uncovered-background pixels between two image frames in a noisy image sequence. Unlike other uncovered-background detection schemes which use motion estimation to distinguish moving from uncovered-background pixels, the Bayes decision test utilizes a single intensity-difference measurement at each pixel and a multiple intensity-difference measurements in the neighborhood of each pixel.

Many image processing and computer vision applications use Bayesian analysis, but it is limited by the computational requirements of extracting data from high-dimensional probability spaces. Recently, new techniques have been developed that enhance Bayesian analysis. A survey of previous and present methods for computing marginal density values for common image models is presented, focusing on implicit polynomial surface models, a Markov random field formulation and parametric polynomial surface models.

The mixture estimation problem is addressed, with applications to unsupervised statistical image segmentation. The notion of a generalized mixture is introduced, and some methods for estimating it are proposed. It is demonstrated that the classical mixture estimation algorithms-expectation-maximization (EM), stochastic EM (SEM), and iterative conditional estimation (ICE) are adaptable. Adaptive versions of SEM, EM and ICE are proposed, in the case of pixel by pixel segmentation.

A general method for estimating the hyperparameter of Gibbs prior from incomplete data is proposed. It is based on a mean-field-like approximation of the Gibbs distribution. Although the method involves substantial computational costs, its practicality is expected to act as sufficient compensation. Experiments indicate that the method can perform well in various conditions when applied to image restoration.

A new approach to using Bayesian motion estimation to interlaced sequences is investigated. The method customizes the manner in which the interpolation is performed to derive an improved motion estimate. Although the study is made in the context of displacement vectors, more precise vectors can be applied. The method has proven to perform better than other techniques on relatively difficult sequences containing high motion, panning and textures.

The use of regularized restoration methods in image processing are discussed. Modelling and analysis of noisy images done by a method using a Bayesian approach is given. The use of Bayesian approach giving various ways for image restoration, which are compared with other regularization method, and a mathematical analysis for calculating noise variance are given.

Research describing the segmentation of synthetic aperture radar (SAR) images is presented. An algorithm for this and segmented video data is developed.

A systematic technique for the segmentation of potential target areas based on a particular wavelet decomposition and a Bayes classifier has been developed. It is concluded that, in general practical cases, the performance of the suggested threshold is often very near the Bayes threshold.

The use of optical diffusion tomography in imaging is examined in terms of Bayesian approaches.