IMDSP-10.1

MOTION-DRIVEN OBJECT SEGMENTATION IN SCALE-SPACE
Ebroul Izquierdo, Mohammed Ghanbari (University of Essex)

In this paper we present a method for motion segmentation, in which accurate grouping of pixels undergoing the same motion is targeted. In the presented technique true object edges are first obtained by combining anisotropic diffusion of the original image with edge detection and contour reconstruction in the inherent scale-space. Contours are then matched according to the distance given by a metric defined on their polygonal approximations and the shape of the one-dimensional intensity function along the contour. Masks of objects are obtained by merging image areas inside of edges having the same motion. The performance of the presented technique has been evaluated by computer simulations.

IMDSP-10.2

Mean Field Decomposition of A Posteriori Probability for MRF-Based Unsupervised Textured Image Segmentation
Hideki Noda (Kyushu Institute of Technology), Mehdi N Shirazi (Osaka Institute of Technology), Bing Zhang (Communications Research Laboratory), Eiji Kawaguchi (Kyushu Institute of Technology)

This paper proposes a Markov random field (MRF) model-based method for unsupervised segmentation of multispectral images consisting of multiple textures. To model such textured images, a hierarchical MRF is used with two layers, the first layer representing an unobservable region image and the second layer representing multiple textures which cover each region. This method uses the Expectation and Maximization (EM) method for model parameter estimation, where in order to overcome the well-noticed computational problem in the expectation step, we approximate the Baum function using mean-field-based decomposition of a posteriori probability. Given provisionally estimated parameters at each iteration in the EM method, a provisional segmentation is carried out using local a posteriori probability of each pixel's region label, which is derived by mean-field-based decomposition of a posteriori probability of the whole region image.

IMDSP-10.3

Energy Matching Based on Deformable Templates
Youssef S Tawfik (Electronics Research Institute), Ahmed M Darwish, Samir I Shaheen (Cairo University)

Snakes, or active contours, have been previously used in computer vision applications to locate and identify objects. However, problems associated with initialization, poor convergence to boundary concavities and high computational complexity, have limited their utility. In this paper, we present a scheme for object localization and classification based on inexact description of the shape. The approach is based on deformable templates and was tested and proven be very robust with respect to changes in object scale, position and orientation as well as noise and local deformations of shape. A coarse-to-fine algorithm was developed to reduce the computational complexity and achieve efficient implementation. Results of applying the algorithm for automatic identification and localization of industrial parts will be presented.

IMDSP-10.4

Object-based Image Segmentation using DWT/RDWT Multiresolution Markov Random Field
Lei Zheng, J. C. Liu, Andrew K. Chan (Texas A&M University), Walter Smith (Stennis Space Center, NRL)

This paper introduces a segmentation algorithm for object-based image coding techniques. This scheme is based on Discrete Wavelet Transform (DWT)/Redundant Discrete Wavelet Transform (RDWT) and Multiresolution Markov Random Field (MMRF). DWT based MMRF works well for images not containing noise. It merges details in the original image with their respective visual objects and divides the image into different meaningful segments according to their textures. The RDWT based MMRF is a generalization of the DWT based MMRF. When the noise level is high, RDWT based MMRF reduces the influence of noise in the segmentation procedure and generates much better results. While at the same time, computation efficiency is reduced. The proposed algorithm has been successfully integrated with our DWT based Region of Interest (ROI) compression coder, the Generalized Self-Similarity Trees (GST) codec, for networking applications.

IMDSP-10.5

Segmentation and Classification of Hand-Drawn Pictograms in Cluttered Scenes - An Integrated Approach
Stefan Mueller, Stefan Eickeler, Christoph Neukirchen, Bernd Winterstein (Department of Computer Science, Faculty of Electrical Engineering, Gerhard-Mercator-University Duisburg)

In this paper, a new approach to identification of handwritten symbols in arbitrary complex environments is presented. 20 different pictograms drawn in different backgrounds can be identified with a recognition accuracy of 90%. In order to perform this challenging task, we use pattern spotting techniques based on pseudo 2-D Hidden Markov Models (P2DHMMs). Practical applications of our approach can be found in many typical mulitmedia document processing tasks, such as localization and recognition of non-rigid objects in image databases, detection of objects in complex scenes, finding trademarks in presence of clutter within videos, processing distorted document images in digital libraries, or content-based image retrieval based on handwritten query symbols.

IMDSP-10.6

Bayesian Estimation for Multiscale Image Segmentation
Srinivas Sista, Rangasami L Kashyap (Purdue University)

We present a solution to the problem of intensity image segmentation using Bayesian estimation in a multiscale set up. Our approach regards the number of regions, the data partition and the parameter vectors that describe the probability densities of the regions as unknowns. We compute their MAP estimates jointly by maximizing their joint posterior probability density given the data. Since the estimation of the number of regions is also included into the Bayesian formulation we have a fully automatic or unsupervised method of segmenting images. An important aspect of our formulation is to consider the data partition as a variable to be estimated. We provide a descent algorithm that starts with an arbitrary initial segmentation of the image when the number of regions is known and iteratively computes the MAP estimates of the data partition and the associated parameter vectors of the probability densities. Our method can incorporate any additional information about a region while assigning its probability density. It can also utilize any available training samples that arise from different regions.

IMDSP-10.7

A Data-Driven Bayesian Sampling Scheme for Unsupervised Image Segmentation
Ed Clark, Anthony Quinn (Trinity College Dublin)

A Bayesian scheme for fully unsupervised still image segmentation is described. The likelihood function is constructed by assuming that the grey level at each pixel site is a realization of a Gaussian random variable of unknown parameters, there being an uncertain number of distinct Gaussian classes in the image. Spatial connectivity between pixel sites is encouraged via a Markov Random Field prior. The task of identifying the model parameters and recovering the underlying class label at each site (i.e. segmentation) is accomplished using a novel reversible jump Markov chain Monte Carlo (MCMC) scheme. This scheme explores the space of possible segmentations via proposals that are driven by the actual image realization---so-called data-driven proposals. The aim is to (i) induce good mixing in regions of high probability, and (ii) optimize the acceptance probability of the proposals. A key development is a stochastic version of a recursive sampling algorithm which has been used in previous work for fast image region splitting. In the current stochastic context, it yields fast and effective split and merge proposals. The performance of the novel MCMC scheme is illustrated in simulation.

IMDSP-10.8

A Spectral 2-D Wold Decomposition Algorithm for Homogeneous Random Fields
Fang Liu (The MIT Lincoln Laboratory), Rosalind W. Picard (The MIT Media Laboratory)

The theory of the 2-D Wold decomposition of homogeneous random fields is effective in image and video analysis, synthesis, and modeling. However, a robust and computationally efficient decomposition algorithm is needed for use of the theory in practical applications. This paper presents a spectral 2-D Wold decomposition algorithm for homogeneous and near homogeneous random fields. The algorithm relies on the intrinsic fundamental-harmonic relationship among Fourier spectral peaks to identify harmonic frequencies, and uses a Hough transformation to detect spectral evanescent components. A local variance based procedure is developed to determine the spectral peak support. Compared to the two other existing methods for Wold decompositions, global thresholding and maximum-likelihood parameter estimation, this algorithm is more robust and flexible for the large variety of natural images, as well as computationally more efficient than the maximum-likelihood method.

< IMDSP-9 IMDSP-11 >

Last Update: February 4, 1999 Ingo Höntsch