Image Quality Assessment Using Author Topic Model

Abstract

In this paper, we propose a novel no reference image quality assessment method. This method performs image quality assessment by incorporating a graphical model. To obtain the results of the image quality assessment, first, we use a set of pristine and distorted images without human subjective scores for training. Second, the images are represented by several quality-aware visual words that are based on natural scene statistic features. Third, author topic model is leveraged to estimate probability of topic for the regions in the test images. At last, the perceptual quality score of the whole image can be obtained by comparing the estimated probabilities of topics with the average distribution of topics for a large number of natural images. Experimental evaluation on the LIVE IQA database demonstrates that the proposed method correlates well with human difference mean opinion scores.