HEVC compressed video visual perception improving method based on GAN

Abstract

The invention provides an HEVC (High Efficiency Video Coding) compressed video visual perception improving method based on a GAN (Generic Area Network). The GAN provides a principle of approaching perception-distortion balance, and the visual perception quality of a compressed video sequence can be improved on the premise of controlling distortion. Therefore, in the work of improving the subjective visual perception quality of the HEVC compressed video, the distance between the distribution of the generated image and the distribution of the natural image is minimized by using the antagonism loss of the GAN, and the GAN generation network can be effectively guided to learn the mapping from the coded frame to the original frame in combination with the video frame after HEVC compression, thereby improving the subjective visual perception quality of the HEVC compressed video. And meanwhile, the mapping is continuously approached through a discrimination network of the GAN. Finally, compared with an SRGAN and an ESRGAN which are used for post-processing of the compressed video, the post-processing network can obtain a better reconstruction visual effect, and objective distortion can be controlled within a certain range.

Description

technical field [0001] The present invention relates to the research of video quality improvement methods in the field of video coding, in particular to a method for improving the visual perception of HEVC compressed video based on GAN. Background technique [0002] As a mainstream video coding standard, HEVC can achieve good compression efficiency, but it will inevitably cause video quality loss during the compression and transmission process. This loss refers to both objective quality distortion and subjective visual blurring. Image restoration algorithms are usually measured by some distortions, such as PSNR, SSIM, IFC, VIF. These distortion metrics are called full reference indicators because the original image is used for evaluation. However, in actual transmission tasks, the original image is not necessarily available, and the reconstruction effect can only be evaluated through the degraded image that needs to be processed. This method of evaluating distortion is calle...

AI Technical Summary

Problems solved by technology

However, in the actual transmission task, the original image is not necessarily available, and the reconstruction effect can only be evaluated through the degraded image that needs to be processed. This method of evaluating distortion is called no reference index, such as the human opinion score that quantifies the visual perception quality

Images