Estimation of Generalized Multisensor Hidden Markov Chains and Unsupervised Image Segmentation
IEEE Transactions on Pattern Analysis and Machine Intelligence
Three-class Markovian segmentation of high-resolution sonar images
Computer Vision and Image Understanding
Simulation and the Monte Carlo Method
Simulation and the Monte Carlo Method
Unsupervised Segmentation Applied on Sonar Images
EMMCVPR '97 Proceedings of the First International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition
Hierarchical statistical models for the fusion of multiresolution image data
ICCV '95 Proceedings of the Fifth International Conference on Computer Vision
Segmentations of Spatio-Temporal Images by Spatio-Temporal Markov Random Field Model
EMMCVPR '01 Proceedings of the Third International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition
| Hi-index | 0.00 |
This paper is concerned with hierarchical Markov Random Field (MRF) models and their applications to multispectral image segmentation. We present an extension of the classic Gaussian model for the modelization of the data likelihood based on a Generalized Gaussian (GG) model, requiring a "shape parameter". In order to obtain an unsupervised multispectral image segmentation, we develop a two step algorithm. In the first step, we estimate the parameters associated with a causal Markovian model (on a quad-tree1) and a generalized Gaussian modeling for the data-driven term, by using an Iterative Conditional Estimation (ICE algorithm [16]). One of the originality of thispa per consists in explicitly decorrelate the multispectral observations during the estimation step on a quad-tree structure. A second step gives the segmentation map obtained with the estimated parameters, according to the Modes of Posterior Marginals (MPM) estimator. The main motivation of the paper is to extend the variety of noise models which results of the distribution mixture on multispectral images. Some results on synthetic and SPOT imagesv alidate our approach.