Texture characterization and image segmentation using Haar patterns and evolutionary processes

Images are composed of objects exhibiting textured surfaces. Anisotropic or irregular textures lacking a deterministic repetitive aspect are difficult to characterize or model. However, such textures may be composed of various textons that can be modelled as specific functions. Starting from this hypothesis, we propose a texture characterization approach based on Haar patterns, combined with an evolutionary approach for the approximate segmentation of the region of interest in color images. The segmentation map is given by the probability of finding a specified pattern and the average distance to discovered solutions and we show that this can be a reasonable estimate of the region of interest. Determining the characteristics of a certain region allows for image segmentation, classification and analysis or further interpretation of image content. In this paper, we show the results we obtain both on synthetic textures and on paintings.