A Multi Agent Approach for Texture Based Classification and Retrieval (MATBCR) using Binary Decision Tree

Texture Analysis has been used in a range of studies for recognizing synthetic and natural textures. We propose a simple, novel and yet effective method for classifying and retrieving images based on texture descriptor. In our system, the information needed for classifying the different types of textures are extracted from the Gabor features, Co-occurrence matrices and Law’s Method. For Feature pre-selection the contextual merit algorithm is used along with the decision tree. For Multi-class classification, SVM with Binary Decision Tree is used. Canberra distance metrics is used for similarity computation. A Multi-Agent system consists of a group of distributed Texture agents that organize their knowledge, goals and plans. In addition, it supports relevance feedback. Our MATBCR model results have been compared with other Texture Based Retrieval System and better prediction accuracy has been observed.