No-reference quality assessment of contrast-distorted images
Min Gang Xu, Zhiming Wang · 2016
Existing quality assessment methods of contrast-distorted images have excellent performance by obtaining information of reference images. However, in actual life, there are not any reference images. To deal with this problem, we propose a simple yet promising no-reference quality assessment algorithm based on the human perception features for contrast-distorted images. First, human visual perception image features are extracted, including perceptual contrast of image, skewness, variance and intensity distribution number of histogram. Then, BP network are utilized to find the mapping function between the feature set and mean opinion score or discrete mean opinion score(MOS/DMOS) given by anthropological observers. Finally, we give the quality scores of test images. Experimental results on CSIQ, TID2008, CID2013 and TID2013 show that our algorithm gives better performance than the state-of-the-art IQA methods.