Quality Enhancement Method For Stereoscopic Video Coding Based On Convolutional Neural Network

With the development of video imaging and multimedia technology,video technology is developing in the direction of ultra-high definition and stereoscopic.To address the efficient compression problems in 3D video,the Joint Collaborative Team on Video Coding proposed the 3D-HEVC standard on the basis of the High Efficient Video Coding(HEVC).However,the standard also degrade the visual quality while minimizing data redundancy.Therefore,how to effectively eliminate distortions and improve the video quality are problems that need to be addreseed.(1)In three-dimensional video system,the texture and depth videos are jointly encoded,and then the Depth Image Based Rendering(DIBR)is utilized to realize view synthesis.However,the compression distortion of texture and depth videos,as well as the disocclusion problem in DIBR degrade the visual quality of the synthesized view.To address this problem,a Two-stream Attention Network(TSAN)-based synthesized view quality enhancement method is proposed for 3D-HEVC in this paper.By learning the global and local information of the synthesized view,the artifacts and holes in the synthesized views can be eliminated.Extensive experimental results show that the proposed synthesized view quality enhancement method achieves significantly better performance than the state-of-the-art methods.(2)The existing neural networks-based synthesized view quality enhancement methods have improved the performance while also increasing the number of parameters.To reduce the number of parameters,save the computational cost and reduce the running time of model,a Residual Distillation Enhanced Network(RDEN)-based lightweight synthesized view quality enhancement method is proposed for 3D-HEVC in this paper.By building a more efficient network structure,the reasoning time can be saved and the parameters of model can be reduced.Finally,extensive experimental results show that the the proposed lightweight method can efficiently enhance the quality of synthesized views and keep the network performacne at a high level.