A No-Reference Image Quality Objective Evaluation Method Based on Multi-Scale Generative Adversarial Network

Abstract

The invention discloses a non-reference image quality objective evaluation method based on a multi-scale generative confrontation network. A similar quality map corresponding to a distorted image is generated through a multi-scale generative confrontation network, and then the similar quality maps of different scales are processed through a convolutional neural network. Regression to get the image quality score. A multi-scale generative confrontation network is trained, and an image quality similarity map is generated for distorted images through a full-reference image quality evaluation method, which is used as the real data set of the discriminant network. Three groups of similar quality maps of different scales are used as datasets, and subjective evaluation scores are used as labels to train the image quality score regression network. The distorted image generates multiple similar quality maps of different scales through a generative network, and then generates an image quality score through an image quality score regression network. The present invention combines the overall distortion degree and local distortion details, thereby further determining the quality score of the distorted image, and reflecting the quality of the distorted image more comprehensively and accurately.

Description

technical field [0001] The invention relates to a non-reference image quality objective evaluation method based on a multi-scale generation confrontation network, which belongs to the technical field of digital image processing. Background technique [0002] Image quality evaluation plays an important role in algorithm analysis and comparison and system performance evaluation in image processing systems. In recent years, with extensive research in the field of digital images, researchers have paid more and more attention to the study of image quality evaluation, and proposed many indicators and methods for image quality evaluation. [0003] From the perspective of whether there is human participation, image quality evaluation methods can be divided into subjective evaluation methods and objective evaluation methods. Subjective evaluation uses people as observers to evaluate images subjectively, and strives to truly reflect human visual perception; objective evaluation metho...

AI Technical Summary

Problems solved by technology

[0008] Aiming at the problem of poor performance of the existing no-reference image quality evaluation, the present invention proposes a no-reference image quality evaluation method, which uses a generative adversarial network to generate multiple similar quality maps of different sizes for a distorted image, and these similar quality maps are passed through Convolutional Neural Networks for Regression to Get No-Reference Image Quality Scores

Images