Mesh Segmentation Based On Face-Face Similarity Probability

Mesh segmentation is a classic topic in computer graphic field and a keyprocess of many3D modelling and graphics applications such as shaperetrieval, deformation, skeleton extraction, shape correspondence, texturemapping, simplification and the popular3D print application recently, etc. Soit is a hot research area and many methods have been developed for meshsegmentation with different advantages and disadvantages depends on thedifferent mesh properties and segmentation criterions, but there is no onealgorithm performs best for every model of all categories. How to integratethe virtues of different segmentation algorithms is an attractive problem. Thispaper proposes a Face-Face Similarity Probability (FFSP) matrix for fusingthe results from different segmentation algorithms. Each element of thematrix represents the probability that two faces belong to the same part. So itis a highly-condensed information package extracted from all the otheralgorithms’ understanding of the mesh model, different from many localgeometric properties, FFSP matrix is a high level and global property of themesh model. After getting the FFSP matrix, we can combine it with theRandom Walks (RW) algorithm for final segmentation by making someadjustments of the main steps of Random Walks algorithm. Compared with classical algorithms, our proposed FFSP-RW method is more competitive inboth quantitative evaluation and the visualized results.