Universal and Unsupervised No-reference Image Quality Assessment Algorithm
Ren Bob · Video Engineering · 2013
To overcome the shortcomings of distortion-specific algorithms and avoid supervised training,a novel no-reference image quality assessment method based on feature pool constructed by visual attention model and edge information is proposed,which is non-distortion-specific and unsupervised. This approach which doesn't limit itself to one or more specific types of distortions can make a good evaluation to all sorts of distorted images,so it is a general-purpose algorithm from this point of view. Besides,the proposed method doesn't have to train with subjective scores,so it is also a truly unsupervised image assessment algorithm. Furthermore,human visual perception characteristics are taken into account when spatial features are extrated from raw-image-patches and assume that regions of interest and edge blocks could affect image quality dramatically. Experimental results show that algorithm performance has good agreements with humans subjective perception.