Feature symbol random field for texture segmentation

This paper defines the feature symbol random field (FSRF) of feature texture images and suggests a novel FSRF-Gibbs model for texture segmentation. The advantages of FSRF are that it can generalize the spatial changed texture feature vectors which come from multichannel analysis, meanwhile, significantly easing the estimation problem of MRF. As a result, finer and more reasonable segmentation is expected by involving both multichannel analysis techniques and a fine Markov random field (MRF) model. A new algorithm is also proposed, which is easy to calculate and yields satisfactory experimental results on Brodatz textures.