A Method of Texture Feature Analysis and Synthesis

Markov random field （MRF） as a basic model for texture synthesis is not good at expressing large-size structures and unstable structures due to its inherent insufficiency on long-range relevancy.In this regard,a novel two-phase framework constructed in pattern space and material space is proposed in this paper to describe the texture structures.By the framework,the basic shape and feature vectors are used to decompose textons’shape,while the feature textures are employed to encode the distribution of textons.The framework takes the advantage both of the flexibility and the accuracy from MRF-like methods.Experiments show that for highly structured textures,our method can generate better results,and also produce certain special effects in synthesis.