基于帧间相关性的动态多帧视频质量增强  

Dynamic multi-frame video quality enhancement based on inter-frame correlation

作　　者：戴天峦 何月顺[1] 何璘琳 陈杰 钟海龙 王文 DAI Tianuan;HE Yueshun;HE Linin;CHEN Jie;ZHONG Haiong;WANG Wen(School of Information Engineering,East China University of Technology,Nanchang 330013,China)

机构地区：[1]东华理工大学信息工程学院,江西南昌330013

出　　处：《传感器与微系统》2023年第9期56-60,共5页Transducer and Microsystem Technologies

基　　金：江西省重点研究计划资助项目(20224BBC41001)。

摘　　要：视频压缩可以减小视频的大小以提高传输效率,但压缩过程中会造成一定程度上的失真和噪声。目前主流的多帧质量增强模型在方法上过于单一,仅使用前后峰值帧或相邻帧进行简单增强,缺乏灵活性。基于上述问题,提出了一种基于帧间相关性的动态多帧质量增强(DMFQE)方法。该方法是采用了两步增强策略,首先,将待增强帧分别输入所设计的光流补偿网络和帧预测网络进行质量增强;然后,将得到的光流增强帧和预测增强帧进行卷积后输入图像重构模块进行重构,获得最终增强帧。实验结果表明:在每一帧上,DMFQE比HEVC/H.265平均多获得0.36 dB PSNR增益;在LDP的编码模式下,DMFQE比单一的多帧质量增强模型MFQE、MFQE2.0拥有更好的有效性。Video compression can be used to reduce video size and improve transmission efficiency.However,the compression process can cause distortion and noise.Current mainstream approach of multi-frame quality enhancement model is too single,as it relies solely on the front and back peak frames or adjacent frames for simple enhancement,lacking flexibility.Based on above problems,a dynamic multi-frame quality enhancement(DMFQE)method based on inter-frame correlation is proposed.The method employs a two-step enhancement strategy.Firstly,the frames to be enhanced are fed into the designed optical flow compensation network and frame prediction network,respectively,for quality enhancement.Subsequently,the enhanced frames obtained from optical flow compensation and prediction enhanced frame are convolved and input into the image reconstruction module for reconstruction,and final enhanced frame is obtained.Experimental results demonstrate that DMFQE achieves an average improvement of 0.36 dB PSNR gain over HEVC/H.265 on each frame.Furthermore,DMFQE exhibits superior effectiveness compared to the single multi-frame quality enhancement models,MFQE and MFQE2.0,in the LDP coding mode.

关 键 词：深度学习 视频压缩 帧间相关性 动态多帧 视频质量增强 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]