基于卷积神经网络的多路视频多视角场景编解码方法  

A Method for Encoding and Decoding Multi Channel Video from Multiple Viewpoints Based on Convolutional Neural Networks

作　　者：于夫 YU Fu(Unit 92941,HULUDAO 123000,China)

机构地区：[1]92941部队,辽宁葫芦岛123000

出　　处：《长江信息通信》2024年第6期85-88,共4页Changjiang Information & Communications

摘　　要：常规的多路视频多视角场景编解码,主要采用视频单帧迭代处理实现编解码过程,忽略了视频中冗余信息对编解码效果的影响,导致编解码结果的视频帧峰值信噪比较低。因此,提出基于卷积神经网络的多路视频多视角场景编解码方法。构建双任务的双残差连接块卷积神经网络,在该网络中最小化视频空间点的距离值,匹配得到视频运动估计矢量特征并补偿,降低冗余信息的影响,在此基础上定义多路视频的编码内容,并通过重构解码帧实现多视角场景的编解码过程。实验结果表明:所提方法应用后得出的视频编解码结果,表现出的视频帧峰值信噪比较高,有效改善了视频质量,满足了多路视频多视角场景的实际应用需求。Conventional multi channel video multi view scene encoding and decoding mainly use single frame iterative processing to achieve the encoding and decoding process,ignoring the im-pact of redundant information in the video on the encoding and decoding effect,resulting in low peak signal-to-noisc ratio of the vidco framcs in the cncoding and dccoding results.Therefore,a multi-channel video multi view scene encoding and decoding method based on convolutional neural networks is proposed.Construct a dual task dual residual connected block convolutional neural network,in which the distance values of video spatial points are minimized,the video motion estimation vector features are matched and compensated to reduce the impact of redun-dant information.Based on this,define the encoding content of multiple videos,and achieve the cncoding and decoding process of multi view scenes by reconstructing and dcoding frames.The cxperimental results show that the video encoding and decoding results obtained after the appli-cation of the proposed method exhibit a high peak signal-to-noise ratio of video frames,effec-tively improving video quality and meeting the practical application requirements of multi view scenarios in multi-channel videos.

关 键 词：多路视频 多视角场景 视频编解码 卷积神经网络 视频处理 编解码方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]