Peak signal-to-noise ratio revisited: Is simple beautiful?

Abstract

Heavy criticism has been directed against using peak signal-to-noise ratio (PSNR) as a full reference quality metric for digitally processed images and video, since many studies have shown a weak correlation between subjective quality scores and the respective PSNR values. In this paper, we show that the low performance of PSNR is often related to a content dependent systematic shift of PSNR values. In scenarios with fixed content and distortion types that are typical for visual communications applications, PSNR may perform closely as well, or in some cases even better than the more complex objective quality models known from the literature. Therefore, the use of PSNR may be justified for comparative quality assessment with fixed content.