Transductive 3D Shape Segmentation Using Sparse Reconstruction

As a three-dimensional model of multimedia data types, have been widely used in many fields of computer-aided design and manufacturing(CAD / CAM),virtual reality, 3D gaming and video animation, following the two-dimensional image, video and sound to drive a new round of multimedia data wave. In recognition of the three-dimensional model, those remarkable shape features easier draw our attention, in order to obtain these remarkable shape characteristics, we must partition the model. This technology is widely applied to computer vision, graphics and other computing applications and research, typical applications include simplified parametric and research grid model, grid model, morphing, 3D reconstruction, 3D model retrieval, GRAPHICS study the texture mapping, mesh reconstruction from point cloud model, and reverse engineering of surface reconstruction. Three-dimensional model can be generally divided into two categories: single segmentation model based segmentation and multi-model consistency. Segmentation refers to the division of a single model based on the model itself and the physical topology of three-dimensional geometry of the surface characteristics of the grid carried. Multi-model consistency segmentation refers to the segmentation model, considering not only the local geometry of the self-possessed, mesh topology, but also for the model database, the same type of model for joint segmentation consistency.Three-dimensional model of the proposed segmentation algorithm is based on multi-model consistency segmentation algorithm as a whole can be divided into three phases: Retrieve a similar model, randomized produce large amounts of segmentation model, sparse reconstruction of the final model segmentation. When a model is divided, first, from the model database to search for the same type of its model, we take the first ten most similar models as a reference model; then, we use the stochastic model segmentation algorithm selected randomly divided, draw the boundaries of hierarchical segmentation approach, smoothing the boundary dividing these models, so you can get a lot of models divided blocks; then, we use the selected reference model for the segmentation block sparse reconstruction, find the nugget The reconstruction error, and finally, we propose a binary integer linear programming algorithm to select from a large number of sub-blocks in the final division of the problem, mathematical modeling as a binary integer linear programming problem seeking optimal value, resulting in a final model segmentation results.