Research On High Efficiency Video Coding And Quality Enhancement Based On 3D-HEVC

With the continuous expansion of video application field and the rapid development of multimedia information technology,the traditional two-dimensional video has been difficult to meet the new needs of the people for better audio-visual in the new era.The three-dimensional video with large perspective and telepresence came into being.It has broad application prospects in the fields of digital film and television entertainment,medical and health care,military and so on.It has gradually become a research hotspot in academia and industry.However,the emergence of three-dimensional video is a double-edged sword.It not only brings people a more infectious and immersive visual feast,but also makes the amount of data grow exponentially and the bandwidth overhead surge.For a long time,seeking the balance between bandwidth cost and high-quality visual experience is the eternal theme of video coding technology.Therefore,in order to reduce the amount of data of three-dimensional video,a more effective video coding scheme is needed.In this context,the three dimensional high efficiency video coding(3D-HEVC)standard came into being.It uses multiview video plus depth(MVD)format for coding,and uses the depth information of depth map to realize the synthesis of virtual views between original views.Although the introduction of depth map alleviates the problem of data surge,3D-HEVC still faces some challenges in practical application.First,the introduction of depth map coding technology leads to the improvement of3D-HEVC coding complexity.Second,the compression coding of three-dimensional video leads to the degradation of the quality of the synthesized virtual view video.These problems have become the key to hinder the popularization and use of3D-HEVC in practical application fields.Therefore,facing the new situation and new challenges,in order to solve the above problems,this thesis takes 3D-HEVC as the research platform.And it takes the depth map rapid intra coding and virtual view video quality enhancement as the starting point.It strives to reduce the 3D-HEVC coding complexity while ensuring the virtual view video quality after the view synthesized process.And then improve the visual quality experience by enhancing the virtual view video quality after the synthesized process.The main research contents and innovative achievements of this thesis are as follows:1.A fast intra coding unit structure decision method for depth map based on multilayer feature fusion is proposed.Aiming at the high computational complexity of depth intra coding for 3D video coding,considering that the features extracted from different convolution layers in convolution neural networks play different roles in image representation,the proposed method is based on the correlation between coding unit structure and its hierarchical features,and takes the realization of rapid decision-making of coding unit structure in depth map as the starting point.A multilayer feature fusion network model for rapid decision-making of coding unit structure is designed.Compared with the 3D-HEVC standard test platform HTM-16.0,the fast decision-making method of depth map intra coding unit structure designed and implemented based on this model can not only ensure the quality of virtual view video after the synthesized process,but also speed up the intra coding process of three-dimensional video,with an average decrease of 37.4% coding complexity.2.A fast intra coding unit depth selection method for depth map based on edge complexity classification is proposed.Aiming at the high computational complexity of depth intra coding for 3D video coding,considering that the combination of HED edge detection network and Otsu threshold segmentation method can be used to obtain the edge region of image significance,the proposed method is based on the correlation between coding unit depth and edge complexity,and aims at realizing the rapid selection of coding unit depth in depth map.An edge complexity classification network model for rapid selection of coding unit depth is designed.Compared with the 3D-HEVC standard test platform HTM-16.0 and the above proposed fast depth map intra coding method,the depth map intra coding unit depth fast selection method designed and implemented based on this model can further reduce the coding complexity of 3D-HEVC on the premise of ensuring the video quality of virtual view after the synthesized process.3.A virtual view quality enhancement method based on generative adversarial networks is proposed.Aiming at the distortion problem of virtual view,by analyzing the characteristics of the virtual view quality after the synthesized process,it is found that the quality of virtual view has been seriously affected after coding,transmission,decoding,and view synthesis.Considering that the generative adversarial networks has good performance in image restoration and image super-resolution,the proposed method designs the process framework of virtual view quality enhancement network model based on the improved generative adversarial networks,and focuses on the construction of virtual view oriented quality enhancement network model.The virtual view quality enhancement method based on this model can effectively improve the quality of synthesized virtual view,with an average increase of 1.127 d B,and achieve good visual effect.