Non-Reference Image Quality Assessment Based on Super-Pixel Segmentation and Information Entropy

Because the existing Blind/Non-reference Image Spatial Quality Evaluator (BRISQUE) algorithm does not take the factor of image information into consideration when evaluating image quality. Therefore, in order to improve the stability and accuracy of the BRISQUE algorithm, a superpixel segmentation blind/non-reference image spatial quality evaluator (S-BRISQUE) based on the information entropy weighting of superpixel segmentation is proposed. First, the image is segmented according to the similarity of the image information, and the information entropy matrix of each part of the sub-image is calculated after the segmentation. Then, the entropy matrix is used to weight the average subtractive contrast normalization (MSCN) parameter matrix, and the MSCN parameter based on the image information is used as the weight. The proposed S-BRISQUE algorithm was verified on the Live benchmark database. Compared with the BRISQUE algorithm and other classic algorithms, the S-BRISQUE algorithm has a higher prediction accuracy rate, also better than some evaluation algorithms.