Co-occurrence-based analysis and synthesis of textures

In the first part of this paper, an algorithm for synthesizing textures obeying a given set of co-occurrence features is presented. It is shown that the synthetically generated images are visually very similar to remotely sensed images coming from the Landsat/TM and ERS1/AMI sensors. This result motivates the use of those features for texture classification. In the second part of the paper, a new algorithm for classifying textures based on co-occurrence feature vectors that are modelled as multi-nomial density functions is presented.