Texture classification and representation by CNN based feature extraction

This paper proposes a novel approach for texture classification by feature extraction based on cellular neural networks (CNNs) and an intelligent arrangement in the design or exact using of the templates in CNN´s. This paper also gives a two-way processing mechanism including the analysis of features extracted from the output of CNNs mapping and a selective training step for obtaining the specific templates in CNNs by genetic algorithms (GAs) for more complicated texture patterns. In this paper, we introduce a one dimensional feature curve as well to indicate the characteristics of original texture patterns from the mapping of the output of CNNs for the latter texture classification. The method introduced in this paper could adaptively choose an appropriate processing procedure for the specific issues towards in specific texture patterns. We finally divide our experiments into two sections, for simple and advanced problems in texture classification. Our experimental results demonstrate the valid template training and at the same time show a satisfactory classification outcome in both defined texture problems.