Perceptual effects of noise in digital video compression

This paper presents the results of subjective viewer assessment of the quality of MPEG-2 compressed video containing wideband Gaussian noise. The video test sequences consisted of seven clips (both classical and new materials) to which noise with a peak signal-to-noise ratio (PSNR) of 28 to 47 dB was added. Software encoding and decoding was performed at five bit rates ranging from 1.8 to 13.9 Mbits/sec. A panel of 32 viewers rates the difference between the noisy input and the compression-processed output. For low noise levels, the subjective data suggests that compression at higher bit rates can actually improve the quality of the output, effectively acting as a low-pass filter. Defining an objective and a subjective measure of scene criticality allows finding the two measures that correlate for the data. For difficult-to-encode material (high criticality), the data suggests that the effects of compression may be less noticeable at mid-level noise. In contrast, for easy-to-encode video (low criticality), the addition of a moderate amount of noise to the input led to lower scores. This suggests that either the compression process may have reduced noise impairments or a form of masking may occur in scenes that have high levels of spatial detail.