Realization And Research Of Several Important Shape Feature Extracting Methods Of 3D Models

We live in a three-dimensional world, and the 3D models are the natureand straight way to describe the object of the real world. Hence the 3Dmodels are widely used in many areas,such as industrial design, electronicgames, virtual reality etc. As the 3D scan and the modeling method becomemore and more popular, there are great many 3D models in the world. Inorder to fully use the models resource, to find the models that one needsquickly and rightly, the research on building the 3D model retrieval systemis becoming active research area. The 3D models shape feature extraction isa important aspect the system. This article is about realization research onseveral important descriptors and the characteristic is that: first of all,theexperiment is on the whole Princeton Shape Benchmark database,comparing with many earlier research in our Country, the result is moreconvincing, and easy to compare with the other production of the researchabroad;The second, as the supporting of the hardware, the voxel is becamemore and more important, our research not only about the mesh but alsoabout the voxel. Although in this article only several descriptors areresearched have been realized, it is easy to realize many other method onthis base, and beginning research on other aspect of 3D model retrieval.The first chapter is introduction part. The background and the sense ofthe research of this article is described. The introduction of the main task ofthis article is the most important part.The second chapter is about the correlative content of shapedescriptor .The shape descriptor is only a part of the system. One needs toknow the whole system if he want to research on shape descriptor moredeeply. This chapter contains 5 parts. In the first part is summarizes themethods of 3D model shape extaction;the second introduce the method forevaluate the descriptor, such as precision-recall plot, the nearest neighbor,the first tier, the second tier;the third part introduce the Princeton ShapeBenchmark database;in the fourth part the two methods of representing the3D models-the voxel and the mesh are introduced. At last,in the fifth part,the necessity of the 3D model predispose. And introduce the method andstep of a common method -Principal Component Analyze.From the third part to the sixth part the article focus on realizing andimproving three important 3D model descriptors and implementation.The third chapter mainly improved the speed and effect in computingthe ray-based descriptor. The descriptor is fist used by Vranic, the basic ideais : project a series of rays alone a certain direction from the center of themodel, every ray may has a point of intersection with the model, andcomputer the distance between the center with the point is a feature of themodel, all the features compose the feature vector. But the computation ofray intersect with the triangle face is so complex that the general way cannotsatisfy the needs.. In application area the speed is also importan .This article,we use the plane coherence of the sampling ray to decrease the rejection test.The ray-based method has been greatly accelerated.The fourth chapter is about the realizing of sphere harmonic transformdescriptor. At first, the article explained the math principle of the sphereharmonic transform. The sphere harmonic transform is a useful method thatcan deal with nearly all the feature vector which is according the definitionof Chebyshev sampling points. By projecting the feature vector from spacedomain to frequency domain, the vector not only can get the rotationinvariance, but also can be insensitive to the small disturb.Considered theprecision and the speed of realize the transform, a free software packagecalled S2kt has been used.The article realize the two ways of sphereharmonic transform method, one is the voxel method, the other is ray-basedmethod. In implementation the voxel, upon the effective method of changethe mesh to surface voxel, a method that filling the inside of the voxelmodel has been designed. The results shows that the solid voxel has a bettereffect than surface voxel in the sphere harmonic transform method.In the fifth chapter, the article has done some primary realizationresearch on the skelecton method. Because of the complex of the 3Dvoxel,the way used in 2D image cannot be easily extended. The article usesanother way, instead of geting a connecting and single wide voxel, get thekey skelecton points first, and design a point connecting arithmetic toconnect the key voxel. When get the key skelecton points, the effective wayof distance transform has been used. At last the article introduce the wayto get the skelecton descriptor and the compare method.Ihe sixth chapter, the article has implemented a visual experimentplatform. The platform contains three parts, the vision part, featureextraction part and the retrieval experiment part. The vision part canexamine the effect of the PCA process and the feature extraction. Thefeature extraction part used to extract the feature of models. More methodsof extraction can be easily embeded. The experiment part is used forexamine the effect of the retrieval. At last on the platform, the severalmethods implement have been compared, and chosen a better realized onour 3D model retrieval system. .