No-reference image quality evaluation method based on self-attention mechanism GAN network

Abstract

The invention discloses a no-reference image quality evaluation method based on a self-attention mechanism GAN network, and particularly relates to the field of image processing, which is composed ofthree parts: a generation network, a discrimination network and a quality evaluation network. The generation network performs feature extraction and recovery reconstruction on an input distorted image, and the discrimination network distinguishes the reconstructed image from an undistorted image as much as possible. According to the method, the adversarial learning intensity is enhanced by addingthe self-attention module and improving the model structure, and a more reliable simulation'reference graph' is output; experiments are carried out on LIVE and TID2013 data sets, results show that theSARAN algorithm provided by the invention is superior to a current mainstream algorithm in overall objective evaluation of quality evaluation of non-reference distorted images, has good performance for different distortion types, and shows that the SARAN algorithm has high correlation with a subjective evaluation result, and better meets the perception consistency of a visual perception system (HVS) on the image quality.

Description

technical field [0001] The present invention relates to the technical field of image processing, and more specifically, the present invention relates to a no-reference image quality evaluation method based on a self-attention mechanism GAN network. Background technique [0002] With the vigorous development of mobile devices and social software, people's requirements for image resolution and clarity are also getting higher and higher. It plays a decisive role in the quality assessment of the distortion or degradation caused in the process of image acquisition and processing. [0003] IQA algorithms can be mainly divided into: full reference, partial reference and no reference IQA algorithms. In recent research, FR-IQA and RR-IQA have achieved a very high correlation with human perception, but they need all or part of the information of the reference image, so their practical application is more limited. In contrast, NR-IQA only takes the evaluated distorted image as input,...

AI Technical Summary

Problems solved by technology

In order to make up for the lack of visual perception difference images, in the Hallucinated-IQA and RAN4IQA algorithms, the GAN network is used to restore the distorted image first, and the restored image and the distorted image are used as input to output the quality score of the distorted image. The existing technology is in It is easy to ignore key details in the image acquisition process, resulting in deviations in image perception and reduced accuracy

Images