Color texture classification based on gravitational collapse

Texture and color are essential attributes to be analyzed for any robust computer vision system. This paper presents a novel method to analyze color-texture images, based on representing states of a simplified gravitational collapse from each image color channel and extracting information from each state using the Bouligand–Minkowski fractal dimension and the lacunarity method. In this approach, we obtained the best classification results when the images of each channel evolved in times t={1,5,10,15}t={1,5,10,15}, each time representing a state, using radius r={3,4,5,6}r={3,4,5,6} for the Bouligand–Minkowski method and box size l={2,3,4,5,6}l={2,3,4,5,6} for the lacunarity method. The best classification results were 99.37% and 96.57% of success rate (percentage of samples correctly classified) for VisTex and USPTex databases, respectively. These results prove that the proposed approach opens a promising source of research in color texture analysis still to be explored.

► In this study, we model a color texture as a simplified gravitational system. ► Each pixel is represented as a particle which interacts with a super massive black hole at the center of the texture. ► We examine changes in the roughness' complexity of each color channel as this system collapses. ► Changes in texture complexity are used to compose a feature vector used in a color texture classification experiment.