Effect of low SNR on visual image quality

摘要

The effect of noise on image quality is investigated using human evaluations of a set of images with a wide range of noise and blur levels. Quality is quantified in terms of the smallest contrast feature that can be reliably identified in a test pattern which can be equated to an effective visual resolution. It is found that the image starts to degrade as SNR falls below 50 and below 10 resolution becomes primarily determined by SNR and is insensitive to blur level. The visual information fidelity metric is shown to accurately model the human response. (C) 2015 Optical Society of America