Research On Simplificaiton And Compression Of 3D Models With Muiltiple Attributes

There is a variety of description forms for three-dimensional(3D)models,such as point clouds and meshes.There are also many attributes for each element of models,such as the position,the normal,the color and the radius for the surfel of point cloud models,and the position,the color or the texture coordinates for the vertex of grid models.Besides,the texture image and the normal image are also very important for 3D grid models.These properties directly affect the surface features of the 3D model.Enabled by the advances in computer graphics technology,high-precision models are being increasingly used in many fields including gaming,filming and scientific simulation.Along with the increasing data come the challenging issue of the compression and simplification of 3D models for more efficient processing.Compression is an effective way to reduce the storage and transmission data of 3D models.In general,different attributes have different spatial distribution.Therefore,different coding and decoding algorithms are proposed according to the characteristics of attributes,which can improve the coding quality and improve the compression efficiency.Since the floating value is usually quantized during encoding,the decoding data may not be the same as the original.Consequently,we need to weigh between the model quality and the limits of the computer environment.In addition to the compression technology,simplification is also an efficient method to reduce the stored,transferred,and processed data of 3D models.Unlike model compression,model simplification does not require coding the model data,but removes redundant information as much as possible while maintaining the original appearance of the model.Therefore,the simplified algorithm is generally lossy and pre-process.A common strategy is to preprocess a model with a series of levels of details(LODs)and select the corresponding LOD according to different application environments.Although some progressive compression algorithms can produce a series of levels of details during coding,the focus of these algorithms is usually on how to compactly code LODs rather than how to improve the quality of simplified models.Although compression and simplification techniques for 3D models have been extensively researched and many related algorithms have been published,these algorithms focus less on the low-bit compression.Further,some of them are limited by the model topology.Although most of these published techniques can make good use of the geometric information of the model(such as the position and the normal attributes),there is little analysis of the texture information.Moreover,most of point cloud simplification algorithms are based on the local heuristic strategy rather than the global optimization.Similarly,earlier 3D mesh model simplification algorithms is also committed to maintain geometric characteristics of models.Later,some simplification algorithms proposed to minimize both the geometric and the textural distortion at the same time,which can retain the appearance of the texture model in a manner.However,these algorithms do not make full use of contents of texture images.In addition,while high-precision texture images may occupy more storage,bandwidth,or computational resources than meshes,the texture image simplification has obtained little concern to researchers.In this paper,we study related works of 3D model compression and simplification algorithms,and propose a new technology,the major contributions of which are as follows:A GLA-based point cloud compression technique is proposed to optimize the level of details and adopt differential prediction coding.The proposed method yields outstanding rate-distortion performance,especially at low bitrates.It literally can process surfaces of arbitrary topology.A point cloud simplification technique is proposed by iteratively selecting and merging local neighborhoods with minimal cost until the desired simplification rate is achieved.Based on the perceptions of human vision,the algorithm uses the local variation of geometric and textural information to calculate the collapse cost of the local neighborhood.Texture differences based on the difference of Gaussian(DoG)are used to enhance texture details of the simplified image.In addition,the algorithm is extended to large-scale model simplification.A novel scheme for collaborative simplification of a 3D triangular mesh and its associated texture image is proposed.The texture image is used to guide the 3D mesh simplification,and the simplified 3D mesh is used to assist the texture image simplification.Firstly,an adapted quadric error metric is proposed to guide the adaptive distribution of 3D primitives on the simplified surface according to both geometric and textural characteristics.Secondly,closed-form optimization is proposed for each replacement vertex to derive its optimal texture coordinates which minimize the local texture deviation on the surface.Thirdly,down-sampling ratios for different texture image regions are optimally determined such that the simplified texture-mapped model appears as close as possible to the original.As a result,the proposed scheme achieves outstanding appearance preservation performance,as experimentally demonstrated.Further,it is generally extensible for other simplification operators with minor adaptations.