In this paper, we have classified previous works and have reviewed 18 representative blur-specific NR-IQA methods. Remarkable progress has been made in the past decades, evidenced by a number of state-of-the-art methods correlating well with subjective evaluations on Gaussian blur images. However, experimental results have also shown that most of the existing methods fail to estimate image quality of realistic blur images. It is the evidence that the blur-specific NR-IQA problem is far from being solved. We have also discussed on realistic blur, especially the issue on image content variation that should be considered in the development of blur-specific NR-IQA methods.
展开▼