Perceptual Sensitivity Based Image Structure-Texture Decomposition

Jinjian Wu, Yuhao Wu, Rong Che, Yongxu Liu

Abstract: Structure-texture decomposition (S-TD) is of significance for many perception-related image processing tasks. In this work, by mimicking the visual perceptual property in the local receptive field of the human visual system (HVS), a novel S-TD method is introduced based on perceptual sensitivity. Considering the perceptual sensitivity of the HVS, three indicators, i.e., the luminance contrast (which measures the change of luminance), the structure anisotropy (which represents the local structure property), and the pattern complexity (which reflects the regularity of the visual content), are firstly defined and measured. And then, according to visual property from the three indicators, image contents are decomposed into five regions, which are the smooth area, the primary edge, the secondary edge, the regular texture, and the irregular texture. Finally, the S-TD method is applied on two perception-related image processing tasks, i.e., image quality assessment and just noticeable difference estimation. And both experimental results verify the effectiveness of the proposed S-TD method.