No-reference image visual quality assessment using nonlinear regression

In this paper, a novel no-reference image visual quality metric is proposed based on fusion of statistical and human visual system based metrics using ε-Support Vector Regression. Different order polynomial regression was also examined as an approximation that has lower computational complexity. Compared to existing image quality assessment metrics, the proposed fused metric is able to better quantify the image quality regardless of the type of degradation. We furthermore improve the image quality assessment by training a separate regression model for each degradation type. The latter degradation specific approach yields near perfect correlation with subjective scores, however, it relies on prior knowledge of the degradation process.