Textural features selection for image classification by Bayesian method

This article proposes an algorithm to optimize the performance in texture classification by Bayesian method. Specifically, we extract several features from the Grey level co-foccurrence matrices (GLCMs) with different distances d and directions θ. We then apply Genetic algorithm to select the suitable features that can minimize the error rate of using the cross validation set. This choice of features continues to be used for classifying test data. Three numerical examples performed with synthetic and real images show the superiority of proposed algorithm over some existing ones. They also present the feasibility and applicability of the proposed method for texture recognition, especially for some practical problems such as material and handwritten digit recognition.