Research On3D Mesh Sequences Compression

In the field of computer graphics, three-dimensional (3D) mesh is the main wayto simulate3D models. In recent years, with the development of3D scanningtechnology and3D modeling technology, the data scale of3D models is growingrapidly. Now, a lot of high-resolution3D models are represented with3D mesheswhich contain thousands of vertices. These3D meshes usually need very large storagespaces, which leads to huge pressure for storing and transmitting of the data.Static3D mesh compression usually utilizes spatial correlation in the3D meshdata. For compressing a dynamic3D mesh sequence, it not only needs to utilizespatial correlation within each frame, but also needs to utilize temporal correlationbetween adjacent frames. If a mesh sequence is topological isomorphic, the interredundance can be removed through mature predicting methods. While, how toremove inter redundance of3D mesh sequences that are not topological isomorphic isa difficult problem. This paper researches on the compression for3D mesh sequenceswhose topology are not isomorphic. The main works are as following three aspects:First, propose a3D mesh sequences compression method based on a thin-platespine(TPS) based surface deformation technique. It uses TPS transformation toimplement the inter prediction of non-isomorphic3D mesh sequences, and uses the7-Zip program to encode I frames and TPS transformational matrices, thus removesintra and inter redundance of3D mesh sequences. Experiments demonstrate that thismethod can achieve good compression quality on the compression of mesh sequenceswhich have small geometrical stretch.Second, propose a3D mesh sequences compression method based on non-rigidregistration. It makes a non-isomorphic mesh sequence becoming isomorphic throughperforming non-rigid registration on the mesh sequence by groups. Consequently, onthe one hand, the inter redundance of topological information can be removed due tothe reason that the meshes in a group are able to share the same topologicalinformation. On the other hand, the intra and inter redundance of geometricalinformation can be removed by encoding difference data of coordinates of verticesbetween adjacent frames in a group. Meanwhile, we learn from the organization formsof Group of Picture(GOP) in video coding technique, support a GOP basedregistration method for3D mesh sequences. We improved the order and number of a target frame by designing different GOP structures. Our compression method canachieve high compression efficiency.Third, propose a geometry images based3D mesh sequences compressionmethod. Geometry image is a concept of representing3D meshes using2D images.This method applies the method based on TPS transformation to geometry imagesequences. It compresses original mesh sequences of geometry image sequencesthrough compressing geometry image sequences by taking TPS transformation,entropy coding directly on pixels of geometry images. Experiments indicate that thecompression efficiency achieved by this method is higher than the compressionefficiency achieved by the method of compressing geometry image sequencesstraightly using traditional image compression algorithms.