An objective measurement tool for MPEG video quality

A two-stage objective measurement model for MPEG-coded video is proposed. The first stage weights the coded video distortion according to the human visual system's response. It computes the frame-by-frame perceptual impairment in the decoded picture with respect to a reference picture; this includes low-pass spatial filtering, a Sobel operation to derive masking coefficients, and spatial masking on the raw error between reference and compressed pictures. The second stage, a cognitive emulator, provides a simulation of human high-level processing of visual information. This includes the very low temporal response of human viewers to image quality changes, and asymmetric behaviour in respect of picture quality changes from bad to good, and vice versa. With this model, we have been able to mimic quite accurately the temporally varying subjective picture quality of video sequences as recorded by the ITU-R SSCQE method.