Video Object Coding And Spatially Scalable Video Coding Using Delaunay Triangulation Mesh

With the rapid development of the third mobile communcation and IP-network,the video service has been one of the most important communication services.In the same time,people expect that the video not only has high code efficiency,but also can be edited,and hope the bit streams can be adapted according to the variations of the available channel capacity in network,the variety of the users'needs and the capacity of users'devices.As is well-known,the block-based video coding usually has visible artifacts under very low bit-rate condition.In order to satisfying all the above needs and overcoming the visible artifacts,this thesis focuses on the research on the object-based and spatially scalable video coding using Delaunay triangulation (DT) mesh.In the research on the object-based video coding, firstly, given the knowledge of object shape, this thesis puts the outside mesh nodes in the object boundary,then the video coding based on DT mesh is expanded to the object-based coding.Also the advantages and disadvantages of this method are analysed.Secondly,with the character of present mesh design algorithm is analysed,a new method is proposed using the image edge information.The results show that this method can place nodes in the most important key points.Also the object-based video coding is extended to the stereo video coding,and is used to synthesize the intermediate view.The result shows that it has a better performance than the block-based method.In the study on spatially scalable coding, a spatially scalable video coding scheme based on DT mesh is proposed in order to offering spatially scalable bit streams. It combines the present inter-layer prediction technique and the mesh-based video coding,and extends the mesh of the base layer to the enhancement layer.Then it uses the motion and texture correlation to improve the coding efficiency of the enhancement layers. Experimental results show that the YPSNR of the proposed techniques decreases about 0.7 dB compared with the single layer mesh-based coding and has 1 dB gain compared with the simulcast.Furthermore,two methods (inter-layer residual prediction and In_scale prediction) which explore simultaneously the temporal and spatial correlation are compared in performance for the scalable video coding. This thesis also proposes a scheme using the inter-layer motion prediction to reduce the time of single layer video coding. It can save about 50% in time.