Content-Based Mesh Design and Occlusion Adaptive GOP Structure

Abstract

2D mesh-based motion compensation can effectively eliminate blocking artifacts that are common in block matching. However, the main difficulty of the scheme is nodal trajectory in occlusion regions. Besides, mesh design is also important because it is related to the efficiency of motion compensation. This work firstly presents a more efficient and delicate algorithm to generate content-based Delaunay triangular mesh (DTM) which is adaptive to temporal activities in a video sequence. Moreover, a mesh-based nodal forward tracking scheme is proposed, in which local mesh structure is locally altered according to occlusion regions. Furthermore, in terms of trajectories of nodes the deformation of the mesh, the change of the content of the video can be perceived, which facilitates the adaptive assignment of GOP structure. The intra-frame of GOB can be assigned at the moment when there is an abrupt change in contents of pictures, e.g. an occluded object (objects) entering picture. The experimental results demonstrate the efficiency of the motion compensation and node tracking and adaptive GOP structure.