Non-reference Image Quality Assessment for Contrast Distortion Based on Pixel Statistics and Color

For most natural images, proper contrast enhancement can achieve better visual quality. However, there are few image quality assessment methods for contrast distortion. We improve a new non-reference image quality assessment model to predict the image quality of contrast changes. Our improvements can be listed in two aspects:1. From the perspective of gray pixel information statistics, we add new perceptual features to the original model, including standard deviation, histogram energy, and skewness. These features enhance the prediction accuracy of the model. 2. Considering the effect of color on the contrast of the image, we extracted two key features related to the overall color of the image, named color saturation and colorfulness. Furthermore, support vector regression (SVR) is used to fuse all features to predict the image quality score, and we achieve better performance on three typical databases (CID2013, CCID2014, and CSIQ).