On the Role of Shaped-Noise Visibility for Post-Compression Image Enhancement

Digital images undergo irreversible compression to enable transmission, a process which induces visible and often annoying distortions. Post-compression enhancement algorithms attempt reduce the visual impacts of these distortions. One such enhancement technique, designed specifically for textures, employs perceptually shaped noise patterns. Although this technique is quite effective, the contrasts of the patterns must be properly scaled, thus necessitating an algorithm that can automatically predict these scaling factors. To help enable such a prediction, we investigated whether patterns which have higher contrast detection thresholds (CTs) [i.e., patterns which require more contrast to be just-visible] also require higher contrast scaling factors (CSs) for proper enhancement. We measured CTs in the current study and compared them with CSs measured in our previous study. Our results support the hypothesis that low CT implies low CS, and vice-versa, but only for those patterns which could markedly improve the visual quality.