Robust approach for textured image clustering

Abstract

Texture is considered as a big issue in image clustering and influence directly on the processing results. So, the idea here is to propose a new approach for image segmentation based on texture-isolation to reduce their effects and to identify easily the different clusters in an image. In this aim, we propose to use first a pre-processing method based on multiscale approach to separate a textured image into texture and geometrical components, applying a filtering process to smooth boundaries, at the end the geometrical component will be used in the classification stage. Several multiscale models, filtering algorithm and classification approaches have been tested on this paper. Simulation results with a comparison study show the good quality of the proposed approach.