Modeling the Perception of Visual Complexity in Texture Images

Abstract

Perception of visual complexity in textures is very important for visual understanding and visual aesthetic evaluation. In this paper, we propose a new model of estimating subjective visual complexity perception of texture images. Compared with the traditional complexity measures based on information theory and fuzzy theory, the proposed model considers human visual perception, and it predicts the visual complexity of a texture corresponding to the subjective visual impression. Multiple linear regression (MLR) is used as a mapping function to map the relationship between the visual complexity perception and five texture characteristics including regularity, roughness, directionality, density and understandability. F-test and correlation analysis are applied to stimulated data and predicted data. The results of F-test (P < 0.01) prove that the proposed model can significantly predict the visual complexity of a texture, and the correlation coefficient between calculated complexity and subjective complexity (r = 0.951) of the testing textures shows that the results predicted by the proposed model are very close to the visual complexity judged by human subjects.