Content-based image quality assessment using semantic information and luminance differences

Abstract

A full-reference image quality assessment (FR-IQA) metric, with emphasis on semantic information changes in different image content areas, is presented. The changes on edge information, that can represent semantic information changes, are calculated based on the characteristics of different image content areas. Considering that edge changes cannot account for luminance changes while luminance changes does affect visual quality of images, the luminance changes are also incorporated into the design of the perceptual quality metric. Experimental results confirm that the proposed metric is consistent with human judgments of quality, and outperforms relevant state-of-the-art metrics across various distortion types.