Full-reference image quality evaluation method based on image distortion types

Abstract

The invention discloses a full-reference image quality evaluation method based on image distortion types. The distortion types of distorted images are predicted by construction of a convolutional neural network model; the quality of the distorted images to be evaluated is evaluated respectively through the four classical full-reference image quality evaluation methods (the PSNR, the SSIM, the JNDand the VIF); corresponding evaluation reference values in four aspects are obtained from each distorted image; evaluation scores obtained by the four different evaluation algorithms are fitted by a nonlinear fitting function; training is carried out according to different distortion types of the images, so that fitting parameters corresponding to each distortion type are obtained; and, for imagesto be detected, corresponding fitting parameters are selected according to the distortion types of the images, so that the final evaluation results of the distortion images are obtained by the fitting function. Compared with the prior art, different theories of image quality evaluation models of the four classical algorithms (the PSNR, the SSIM, the JND and the VIF) are fused; the advantages of each model are exerted; and thus, the performance of the new method is better.

Description

technical field [0001] The invention relates to various fields such as image processing, machine learning, and image quality evaluation, and in particular relates to a full-reference image quality evaluation method based on image distortion types. Background technique [0002] Image quality assessment is an important research field in image processing. Image quality evaluation methods are divided into subjective evaluation and objective evaluation. Among them, subjective evaluation is the most true reflection of human eyes on image quality, and is the most reliable evaluation index. However, subjective evaluation consumes a lot of manpower and time, and cannot be applied to real-time image quality evaluation. The objective image quality evaluation algorithm can be better embedded in the system, can realize real-time evaluation and low cost, and is suitable for practical applications. Therefore, it is of great significance to develop an objective image quality evaluation m...

AI Technical Summary

Problems solved by technology

[0004] The article "Gradient magnitude similarity deviation: A highly efficient perceptual image quality index" records the tests of 11 currently the best objective image quality evaluation algorithms. The test results show that: for a single distortion type, the performance of most algorithms is very good. OK, but overall the performance of all the algorithms is not great for an image library with multiple distortion types

The reason may be that the human visual system is a multi-scale system, and there are different evaluation strategies for different situations (such as distortion types), so the evaluation model of a single strategy cannot be highly consistent with human subjective evaluation

Images