Comments on objective quality assessment of JPEG images with visible differences

Abstract

We analyse the results of a comprehensive collection of objective quality metrics for high definition compressed visual content. The investigation provides guidelines that allow a reliable JPEG image quality assessment, and therefore the development of perceptual-based image/video coding standards. We show on a comprehensive, recent subjective dataset with visible differences that visual information fidelity, most apparent distortion, and gradient-based objective measures have reached the best agreement with the subjective quality scores, where contrast is preferred according to the luminance and structure changes.