Author(s)

Yao Ma, Fuming Sun, Shijie Hao

Corresponding Author

Yao Ma

ABSTRACT

Due to the serious shortage of training data, CNN's research on non-reference image quality assessment (NR-IQA) is very constrained. In this paper, the existing neural network research on NR-IQA is summarized, and the methods to solve the problem are divided into three types: image segmentation, pre-training migration and unsupervised sequence learning. Then from these three aspects, a more representative algorithm is selected to test and compare performance on different datasets, and analyze the advantages and problems of the three. In the end, the direction of NR-IQA development is discussed, which provides a comprehensive reference for researchers in this field.

KEYWORDS

CNN, image segmentation, pre-training migration, unsupervised sequence learning, NR-IQA