No-reference image contrast quality assessment of new media video based on generated perceptual difference

Abstract

The reference quality evaluation method is constrained by the reference image, and the reference image cannot be obtained in practical application. The new media video picture contrast no reference quality evaluation model based on the generated perceptual difference is designed to provide a scientific basis for improving the new media video picture quality. Taking the video image to be evaluated and the given gradient difference image as an input, a visual perceptual difference image is generated through the confrontation learning of the generation network and the discrimination network in the generation of the countermeasure network; The regression network of quality assessment is trained with the quality score of the image to be evaluated and its true contrast as input; The generated perceptual difference image and the image to be evaluated are input to the trained quality assessment regression network, to obtain the contrast quality assessment score output of the image to be evaluated. The results show that the model can generate visual perceptual difference images of different video images, so as to master the contrast distortion characteristics of different images, and complete the contrast quality assessment of different images based on this. The results obtained are consistent with the subjective assessment scores, the average values of SROCC, PLCC and RMSE in the evaluation results of this model are 0.9804, 0.9749 and 3.1266, respectively. the overall assessment results are accurate, and the assessment performance is stable and reliable.