An approach for learning the optimal “tuned” masks based on differential evolution algorithm

Texture image classification is a significant topic in many applications of machine vision and image analysis. The texture feature extracted from the original image by using the “Tuned” mask is one of the simplest and most effective methods. However, the primary gradient based training method almost always falls into the local optimum which might be improved through some commonly used evolutionary algorithms, such as genetic algorithm (GA) and particle swarm optimization (PSO). Unfortunately, these algorithms will easily trap into the local optimum as well. For the sake of learning “Tuned” mask with the better performance, this paper propose to employ differential evolution algorithm to generate the optimal “Tuned” mask. Experiments on some texture images from the Brodatz album show that the “Tuned” mask training method proposed in this paper is very effective for classifying texture images and outperforms the “Tuned” mask training method based on genetic algorithm and particle swarm optimization algorithm.