PSNR-HVS is a perceptually weighted variant of PSNR that keeps the decibel scale and speed of ordinary PSNR but weights the pixel errors by a model of the human visual system — specifically its contrast sensitivity — so errors the eye barely perceives count for less than errors it notices. It was built to fix PSNR's core weakness: the lesson of PSNR is that the problem was never the decibel scale but treating all pixel errors as equal, and weighting them by how the eye works makes a PSNR-style metric useful again. A stronger relative, PSNR-HVS-M, adds visual masking — the way busy, textured regions hide errors — and in its 2007 evaluation tracked human judgement considerably better than both PSNR and SSIM. These variants sit between raw PSNR and the heavier perception-trained metrics: faster than VMAF, closer to the eye than plain PSNR, useful when you want PSNR's simplicity with a number nearer perception. They remain full-reference and inherit the alignment sensitivity of any pixel-based metric.
