Comparative analysis of universal methods no reference quality assessment of digital images

Abstract

The main purpose of this article is to conduct a comparative study of two well-known no-reference image quality assessment algorithms BRISQUE and NIQE in order to analyze the relationship between subjective and quantitative assessments of image quality. As experimental data, we used images with artificially created distortions and mean expert assessments of their quality from the public databases TID2013, CISQ and LIVE. Image quality scores were calculated using the NIQE, BRISQUE functions and their average. The correlation coefficients of Pearson, Spearman and Kendall were analyzed between expert visual assessments and quantitative scores of the image quality, as well as between the values of three compared indicators. For the experiments, the Matlab system and values of its functions niqe and brisque normalized to the range [0, 1] were used. The computation time of niqe is slightly less. The investigated functions poorly estimate the contrast of images, but the additive Gaussian noise, Gaussian blur and loss in compression by the JPEG2000 algorithm are better. The BRISQUE measure shows slightly better results when evaluating images with additive Gaussian noise, while NIQE for blurred by Gaussian. The average of the normalized values of NIQE and BRISQUE is a good compromise. The results of this work may be of interest for the practical implementations of digital image analysis.