No-reference image quality assessment based on regional mutual information
Yanjie Wang, Yanjie Wang, Jianjun Zhou, Li Jia, Yue Wang, Yue Wang · 2015
We study the regional mutual information to build a powerful and effective no-reference (NR) image quality assessment (IQA) approach.This approach does not need to compute distortion-specific features, but use the features which can distinguish the distortion in images across the distortion types.Such features are extracted from the image regional mutual information, which includes modified regional mutual information and a novel way to describe such information via wavelet transform.Operating within the 2-step framework, these features are tested on the LIVE database and show a nice correlation with human subjective opinions of image quality.