Blind Quality Assessment for Multiply-Distorted Images via Distortion Decomposition

In the real world, digital images are often contaminated by several distortions simultaneously, and the inability to identify the types of these distortions and describe their joint effects makes quality assessment of multiply-distorted images an intractable problem. In this paper, a new no-reference image quality assessment (NR-IQA) approach is proposed by summarizing various distortions into detail loss (DL) and detail redundancy (DR), according to the visual impact caused by the changes of image details. DR is expressed by the features of the absolute error map between the distorted image and its denoised version, while DL is represented by the difference between the denoised image features and the estimated pristine features. Experiment results on multiply-distorted image databases demonstrate that the proposed metric is superior to existing NR-IQA methods in terms of the coherence with human subjective rating.