Model Fitting and Model Evidence for Multiscale Image Texture Analysis

Mihai Datcu
2004 AIP Conference Proceedings  
This paper, begins with an overview of the two levels of Bayesian inference: model fitting and model selection, and shows how they can be used for the image texture analysis. The applied models are the Gauss-Markov and Gibbs auto-binomial Random Fields. In the second part the article introduces a linear model for the image wavelet coefficients able to explain the full description of the spatial, inter-scale and inter-band behavior of a multi-resolution decomposed image. The model parametrs,
more » ... l variance and evidence are used to chraterize the image texture. INRODUCTION Images have high diversity information content, thus their modelling and information extraction are not easy tasks. The texture is one of the most important image feature, both as objective signal characterization and at the subjective visual perception level. During the last decades, among many approaches to deal with the texture analysis, important achievements have been obtained by using Bayesian estimation methods mainly for Gibbs-Markov random fields models GMRF. The complexity of textures and their multi-scale behavior, was studied recently either using libraries of GMRF models of various orders [1], or performing the analysis in the multi-resolution wavelet transformed domain [2]. This paper, in the first part, begins with an overview of the application of the two levels of Bayesian inference, model fitting and model selection, for the texture parameters estimation and the selection of model order. The used models are the Gauss-Markov and Gibbs auto-binomial Random Fields. FIGURE 6. Classification results. Clockwise: original images, Gauss-Markov, auto-binomial, and wavelet based models.
doi:10.1063/1.1835195 fatcat:sfnmeqrr5zajxapfmioj4zwd7m