A review on image segmentation techniques

Dynamic Bayesian Networks: Representation, Inference and Learning by Kevin Patrick Murphy Doctor of Philosophy in Computer Science University of California, Berkeley Professor Stuart Russell, Chair Modelling sequential data is important in many areas of science and engineering. Hidden Markov models (HMMs) and Kalman filter models (KFMs) are popular for this because they are simple and flexible. For example, HMMs have been used for speech recognition and bio-sequence analysis, and KFMs have been used for problems ranging from tracking planes and missiles to predicting the economy. However, HMMs and KFMs are limited in their “expressive power”. Dynamic Bayesian Networks (DBNs) generalize HMMs by allowing the state space to be represented in factored form, instead of as a single discrete random variable. DBNs generalize KFMs by allowing arbitrary probability distributions, not just (unimodal) linear-Gaussian. In this thesis, I will discuss how to represent many different kinds of models as DBNs, how to perform exact and approximate inference in DBNs, and how to learn DBN models from sequential data. In particular, the main novel technical contributions of this thesis are as follows: a way of representing Hierarchical HMMs as DBNs, which enables inference to be done in O(T ) time instead of O(T ), where T is the length of the sequence; an exact smoothing algorithm that takes O(log T ) space instead of O(T ); a simple way of using the junction tree algorithm for online inference in DBNs; new complexity bounds on exact online inference in DBNs; a new deterministic approximate inference algorithm called factored frontier; an analysis of the relationship between the BK algorithm and loopy belief propagation; a way of applying Rao-Blackwellised particle filtering to DBNs in general, and the SLAM (simultaneous localization and mapping) problem in particular; a way of extending the structural EM algorithm to DBNs; and a variety of different applications of DBNs. However, perhaps the main value of the thesis is its catholic presentation of the field of sequential data modelling.

/pdf/dynamic-bayesian-networks-representation-inference-and-yy4fhfxdcj.pdf

Dynamic bayesian networks: representation, inference and learning

A scheme is developed for classifying the types of motion perceived by a humanlike robot. It is assumed that the robot receives visual images of the scene using a perspective system model. Equations, theorems, concepts, clues, etc., relating the objects, their positions, and their motion to their images on the focal plane are presented. >

http://ieeexplore.ieee.org/iel2/815/3148/00100651.pdf

Robot vision

A computational approach for analyzing visible textures is described. Textures are modeled as irradiance patterns containing a limited range of spatial frequencies, where mutually distinct textures differ significantly in their dominant characterizing frequencies. By encoding images into multiple narrow spatial frequency and orientation channels, the slowly varying channel envelopes (amplitude and phase) are used to segregate textural regions of different spatial frequency, orientation, or phase characteristics. Thus, an interpretation of image texture as a region code, or carrier of region information, is emphasized. The channel filters used, known as the two-dimensional Gabor functions, are useful for these purposes in several senses: they have tunable orientation and radial frequency bandwidths and tunable center frequencies, and they optimally achieve joint resolution in space and in spatial frequency. By comparing the channel amplitude responses, one can detect boundaries between textures. Locating large variations in the channel phase responses allows discontinuities in the texture phase to be detected. Examples are given of both types of texture processing using a variety of real and synthetic textures. >

Multichannel texture analysis using localized spatial filters

Digital processing of speech signals

Bayesian networks provide a natural, concise knowledge representation method for building knowledge-based systems under uncertainty. We consider domains representable by general but sparse networks and characterized by incremental evidence where the probabilistic knowledge can be captured once and used for multiple cases. Current Bayesian net representations do not consider structure in the domain and lump all variables into a homogeneous network. In practice, one often directs attention to only part of the network within a period of time; i.e., there is “localization” of queries and evidence. In such case, propagating evidence through a homogeneous network is inefficient since the entire network has to be updated each time. This paper derives reasonable constraints, which can often be easily satisfied, that enable a natural {localization preserving) partition of a domain and its representation by separate Bayesian subnets. The subnets are transformed into a set of permanent junction trees such that evidential reasoning takes place at only one of them at a time; and marginal probabilities obtained are identical to those that would be obtained from the homogeneous network. We show how to swap in a new junction tree, and absorb previously acquired evidence. Although the overall system can be large, computational requirements are governed by the size of one junction tree.

/pdf/multiply-sectioned-bayesian-networks-and-junction-forests-55e0yw04lb.pdf

Multiply sectioned bayesian networks and junction forests for large knowledge-based systems

This paper comments on the optimality of the Laplacian of a Gaussian edge detection filter which localizes edges through zero crossings in the filtered image. The arguments of both Marr and Hildreth, and Dickey and Shanmugam are reviewed to establish that the filter is optimal in the sense of maximizing output image energy near edge features. This filter's principal advantage over other edge detectors is that its response is user-adjustable through selection of a single parameter, the Gaussian standard deviation. However, no clear method for the selection of this parameter has been provided. The problem is addressed here by applying the filter to two ideal periodic edge models blurred by a Gaussian distribution point-spread function. The observed response to the edge spacing and blur standard deviation is then translated into a filter parameter design procedure. The problems of optimum filter performance in the presence of additive Gaussian noise are then addressed. The problem of selecting the sampled filter's coefficient word size is dealt with in a companion paper.

Optimal Edge Detector Design I: Parameter Selection and Noise Effects

Paper: Multiply sectioned Bayesian networks for neuromuscular diagnosis

A companion paper describes the design and implementation of the Laplacian of a Gaussian edge detection filter which localizes edges through zero crossings in the filtered image. Accuracy in the presence of noise has been found to be proportional to the square root of the filter's standard deviation. Digital implementation of any continuous filter requires sampling and coefficient quantization. The sampled filter was examined, but a method is proposed here for selection of a minimum coefficient word size for direct-form implementation to satisfy in-band rejection bounds.

Optimal Edge Detector Design II: Coefficient Quantization

Previously, estimating vocal-tract filters and glottal waves from vowel sounds imposed either the invalid assumption that glottal waves over closed glottal intervals are zero, or parametric models for glottal waves, resulting in biased vocal-tract-filter estimates and glottal-wave estimates lacking information over closed glottal intervals. We obtain unbiased vocal-tract-filter estimates from sustained vowel sounds, for which the glottal waveforms are periodically stationary random processes. Two equations are constructed each relating the vocal-tract filter to the sound signal and the glottal wave over one of two closed glottal intervals. By subtracting one equation from the other, the periodic components of the glottal wave are eliminated from the vocal-tract-filter estimation, and an unbiased vocal-tract-filter estimate is obtained. The average of many such estimates from different closed glottal intervals of the sound is the final estimate, which is used to obtain the glottal wave by inverse filtering the sound. The results obtained from vowel sounds /a/ produced by some subjects are presented. Over closed glottal phases, the glottal waves obtained are nonzero. During vocal-fold colliding, they increase; during vocal-fold parting, they decrease or even increase. The vocal-tract filters obtained yield vocal-tract area functions similar to that measured from an unknown subject's magnetic resonance image.

Michael P. Beddoes

Papers

Multiply sectioned bayesian networks and junction forests for large knowledge-based systems

Optimal Edge Detector Design I: Parameter Selection and Noise Effects

Paper: Multiply sectioned Bayesian networks for neuromuscular diagnosis

Optimal Edge Detector Design II: Coefficient Quantization

A New Method for Obtaining Accurate Estimates of Vocal-Tract Filters and Glottal Waves From Vowel Sounds