No-reference/Blind Image Quality Assessment: A Survey

Abstract

In recent years, no-reference/blind image quality assessment (NR-IQA), as a fundamental but challenging research problem, has been attracting significant attention in the field of digital image processing. NR-IQA aims to build a computational model to quantitatively predict the subjective quality from the distorted image itself without any reference image. Although great efforts have been employed to develop various NR-IQA algorithms, due to its intrinsic difficulty, the issues of NR-IQA are still challenging and remain largely unexplored to date. In this paper, we comprehensively review the fundamental developments of NR-IQA with more emphasis on general-purpose NR-IQA algorithms. First, a brief introduction about NR-IQA algorithms is given. Afterwards, distortion-specific and general-purpose NR-IQA algorithms are presented, focusing on the two major aspects of features extraction and quality prediction. The performances of state-of-the-art NR-IQA algorithms on three commonly used benchmark databases are also evaluated. As the field of NR-IQA is still growing rapidly, we highlight some open challenges about the developments of NR-IQA with some further discussions. We hope that the review presented in this paper can provide a good insight into recent developments in NR-IQA for the researchers who are interested in NR-IQA.