Research On Image Quality Assessment Based On Subjective Perception

Image quality assessment is a very important topic in image and video coding, because it is not only a ruler for performance evaluation of coding algorithms but also a metric for ratio-distortion optimization in coding.Because of the heavy dependence of Neurophysiology, image quality assessment based on subjective perception has been a very controversial issue. After collecting and studying many materials of Neurophysiology, the paper analysis the procedure of visual information and propose a novel image quality assessment according to the Human Visual System.Inspired by the fact that different spatial frequencies have different influence on the human visual perceive and further more, local and global information effect the visual sensitivity in different way, the paper proposes an image quality metric based on global and local information (namely Spatial Frequency Component with Global and Local information SFCGL). Based on the fact that the different human visual sensitivity of colors and multistage of visual information procedure, the paper did some further research on SFCGL and proposes Multi-scale-SFCGL and color image quality assessment COR_SFCGL. In SFCGL, the distortion of the global information mostly existing in low spatial frequency is measured by a rectified mean absolute difference metric, and the distortion of the local information mostly existing in high spatial frequency is measured by Structure Similarity (SSIM). These two measurements are combined using a newly proposed abruptness weighting that describes the uniformity of the residual image. In Multi-scale-SFCGL, after decomposing the reference and distortion images by Haar wavelet, we evaluate the distortion is same way in cross-high and low spatial subbands as in SFCGL, as in the vertical-high and horizontal-high spatial subbands, the metric uses the correlation to measure the distortion of corresponding vector. The result is fused by the linear weighted composition. The color image quality assessment is an added research in the paper. Experimental results on LIVE database show that the proposed metric outperforms the SSIM and achieves performance competitive with the state-of-the-art metrics Visual Information Fidelity (VIF), but the improve edition Multi-scale-SFCGL performs better than all the other image quality metrics.