Objective evaluation method for full reference image quality based on neural network learning integration

Abstract

The invention discloses an objective evaluation method for the full reference image quality, which comprises the steps of applying a BP neural network to image quality evaluation, designing a visual multi-channel multi-algorithm adaptive integration BP neural network image quality prediction model, inputting a distorted image into the BP neural network based on visual multi-channel evaluation results of various objective evaluation algorithms, performing supervised learning and training on the BP neural network by taking the score of a human eye subjective test result as a training objective,then predicting and outputting objective evaluation results of the various objective evaluation algorithms, and performing adaptive integration on the objective evaluation results of the various algorithms to obtain final objective evaluation for the quality of the distorted image. The method disclosed by the invention comprehensively improves the level of various indexes of the evaluation methodsuch as PSNR, SSIM and SVD, exceeds the latest evaluation methods such as visual feature perception processing and visual psychological derivation integration, and has better evaluation stability.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a full-reference image quality objective evaluation method based on neural network learning fusion. Background technique [0002] As a widely used signal, image plays an important role in various fields such as information acquisition, transmission and processing. At present, with the improvement of cloud computing capabilities and the rise of artificial intelligence research, various application services based on image terminal processing platforms have achieved unprecedented development. However, image signals are easily polluted, so the research on image quality evaluation is of great significance. In the field of image quality evaluation research, the objective method has become a research hotspot in this field because of its automatic and continuous efficient work mode. Among them, the research significance of full-reference image quality evaluation is p...

AI Technical Summary

Problems solved by technology

In recent years, integrating visual system characteristics into image quality evaluation has become a research hotspot, for example, significant distortion MAD, feature similarity FSIM, visual saliency VSI, etc., but these methods have two problems. One problem is that image feature processing algorithms lack visual Theoretical basis of characteristics, which makes its evaluation performance unstable; another more prominent problem is the multi-channel characteristics of subjective vision of the human eye. For different objective evaluation algorithms, there are different subjective and objective mapping relationships. For example, for each visual The frequency sensitivity characteristics of the channel, the above methods all use a unified contrast sensitive function, which reduces the performance of various objective evaluation methods

Internal Generative Mechanism (IGM) is based on the free energy field theory of the brain, and evaluates image quality through information perception maximization algorithms, but its information processing algorithm is too simple to reveal the working mechanism of the visual brain, and it also lacks powerful Experimental results confirmed

Images