Full-Reference Image Quality Assessment Procedure Based on Rice Distribution Model

Abstract

The problem of full-reference image quality assessment is considered based on the application of the mathematical model of the Rice distribution. The gradient field of an image is adequately described by the Weibull distribution, which allows one to effectively analyze image properties, evaluate their similarity, classify them by quality, etc. In this paper, an attempt is made to solve similar problems using the above-mentioned model, relying, in particular, on additional properties of the Rice distribution associated with the normal approximation of the latter. It is shown that the structural similarity measure used in different problems is also applicable to the case of the Rice gradient field model. In particular, images from the TID2013 database are experimentally studied. The modeling results obtained from both the Weibull and Rice distribution models were compared using the mean square and structural similarity measures, as well as the Mean Opinion Score (MOS) values. It is shown that the types of distortions in these indicators are in complete agreement, while for some other types, the Rice distribution model shows better results.