Markov random fields for recognizing textures modeled by feature vectors
This paper decribes a new probabilistic framework for recognizing textures in images. Images are described by local affine-invariant descriptors and by spatial relationships between these descriptors. We propose to introduce the use of statistical parametric models of the dependence between descriptors. Hidden Markov Models (HMM) are investigated for such a task using recent estimation procedures based on the mean field principle to perform the non trivial parameter estimation they require. Preliminary experiments obtained with 140 images of seven different natural textures show promising results.