An Efficient Video Coding Method Based On Video Colorization And Quality Enhancement

With the rapid development of imaging and display technologies,the high-definition video is widely popularized and spread in daily life.Compared with H.264/AVC,which is the previous generation video coding standard,High Efficient Video Coding(HEVC)can double the compression efficiency with the same video quality.However,the existing video coding standards are hard to provide higher video quality under the limited storage and bandwidth resources due to the increase of video resolution,frame rate and sampling precision.Therefore,the compression efficiency of HEVC needs to be further improved.In addition,the quality of the compressed video needs to be enhanced due to the recognizable distortion that introduced in the compressed video during the acquisition and compression process.(1)Traditional video coding standards directly compress the luminance and chrominance components of the raw video without any preprocessing.According to the statistical analysis,there is a strong correlation between the coding parameters of the luminance and chrominance components.Therefore,a novel colorization-based video encoding method is proposed in this paper.The reconstructed luminance component and the trained colorization model are sent to the decoder,by which the chrominance component can be reconstructed in the decoder.To accurately recover the chrominance component,a generative adversarial network(GAN)model is adopted in the decoder.Experimental results show that the proposed colorization-based video coding method can further compress video without affecting the visual quality of video.(2)There are different types of distortions in the video after the acquisition and compression process.In order to improve the quality of the video,this paper proposes a GANbased video image quality enhancement method to post-process the video at the decoder of HEVC.The method consists of a multi-module cascaded generative network and an adaptive multi-scale discriminative network.In addition,this paper proposes a novel loss named contrast loss,and on this basis,establishes a new mixed loss function.Experimental results show that the proposed enhancement method can both improve the quality of image and video.