Affective Property Computation of Visual Texture

Abstract

Affective computing of visual textures is a cross-disciplinary research field. In this paper, we propose a hierarchical feed-forward layer model represented by multiple linear regression to investigate the relationship between human aesthetic texture perception and computational low-level texture features. Instead of black-box models not allowing any interpretable insights, we tried to build white-box models within each layer that can be psychologically interpreted from aspects of both, structure and interrelations between aesthetic properties and texture features. Based on these combined with the hierarchical structure, someone can gain the degree of influence of texture features as well as properties in lower layers on to the properties in higher layers, achieving a kind of step-wise psychological interpretation in terms stage-wise cognitive depth.