Robust Feature Extraction from Texture Based on Multi-dimensional Morphology and Its Application

Abstract

In this paper, we propose a robust feature extraction method for color texture image based on multi-dimensional morphology. Multi-value co-occurrence matrix of multi-value morphology pattern spectrum of skeletons is used to represent texture structure. Color features are also extracted from RGB components based on similar binary morphology pattern spectrum. The extracted feature vectors have robustness to geometric distortions such as shift, rotation, scaling, skew, projection and their combinations. In simulations, robustness performance of the proposed method is verified to show the effectiveness under various distortion conditions. The effectiveness is examined by applying to image retrieval and classification systems. Accuracy of retrieval and classification is evaluated compared with those of HLAC based approach.