3D Mesh Parallel Segmentation Using Mean Shift

With the increasing scale of3D mesh model of data, high order digital geometry calculation is becoming more and more complex, and the real-time of digital geometry processing is growing. The emergence of the GPU greatly improved the calculation of the relevant work efficiency, and it decreased the video card to CPU dependence on most of the graphics processing work from the CPU transfer to the GPU. At the same time the GPU high performance, high precision of parallel computing of digital geometry processing work to provide strong support.3D mesh model of segmentation is one of the important basic research problem in the digital geometric processing, It has a wide range of applications, for example, the mesh simplification, compression and editing, texture mapping, shape matching, parametric and animation deformation and so on, through the three dimensional model of the division, it reduced the complexity of the algorithms. The international domestic proposed3D segmentation algorithms are existed segmentation, segmentation boundary meaningless and so on, and it has other defects in the operation speed and precision segmentation.The existing3D mesh segmentation algorithm most derived from image segmentation algorithm, such as watershed algorithm, k-means clustering algorithm, hierarchical clustering algorithm, pedigree clustering algorithm and so on, in the three dimensional geometry space partition problems all have very good promotion application. Mean Shift algorithm is a nuclear density gradient no parameters estimation method, the method is simple, without the pretreatment, good real-time, treatment goals more robust deformation. It has a better performance in many applications, for example, video, video signal in tracking analysis, image segmentation and so on.Base on serial of Mean Shift segmentation algorithm and high performance parallel operation characteristics of GPU, this paper proposes an new3D mesh model Mean Shift parallel segmentation algorithm. It overcomes traditional serial Mean Shift algorithm complexity, high speed slow faults. This method firstly based on discrete surveying the distance of the local extremum3D mesh model extracted the salient features of the point, and then from the obvious characteristics executing Mean Shift algorithm. For local curvature distribution is more complicated and excessive segmentation problem, this paper realized the post-processing, and the quality of the division has certain fast segmentation result. Compared with other similar algorithms, this method enhanced the efficiency of the algorithm, and improved the segmentation effect.