No-reference image quality assessment based on natural scene statistics in RGB color space

There are strong correlations between the color components in the RGB color space, and distorted images can change those correlations. Based on this, a novel general-purpose no-reference image quality assessment (NR-IQA) method is proposed. Firstly, according to the color perception characteristic that human vision is more sensitive to the green component in the RGB color space, the statistical features of MSCN coefficient and its four neighboring coefficients of the G component are extracted. Secondly, on the basis of the correlation analysis between R, G and B components in RGB color space, the mutual information between the color components in RGB color space, their textures and their phases are calculated respectively. The statistical features of mutual information are used to describe the correlation between the color components in the RGB color space. Moreover, based on the aforementioned statistical features, support vector regression (SVR) and support vector classifier (SVC) are used to construct a NR-IQA model and image distortion type recognition model, respectively. At last, in order to analyze the correlation with different mean opinion score (DMOS), classification accuracy and computational complexity, a large number of simulation experiments are carried out in the LIVE, CSIQ and TID2008 image quality evaluation databases. Simulation results show that this method is suitable for many common distortions and consistent with subjective assessment, and that the Spearman's rank ordered correlation coefficient (SROCC) and the Pearson's linear correlation coefficient (PLCC) in LIVE image quality evaluation database are more than 0.942.In addition, the recognition accuracy of the recognition model is up to 93.59% and significantly superior to all present-day distortion-generic NR-IQA methods. Copyright © 2015 Acta Automatica Sinica. All rights reserved.