Content-based 3D Model Retrieval

With the rapid development and extensive use of new scanners and interactive modeling tools, the number of 3D models increases quickly recent years. 3D model plays a more and more important part in many fields and many large 3D model repositories come forth. The need for retrieval of models from large repositories of 3D shapes has gained prominence in the graphic and vision communities. This paper focusd on feature extraction algorithm based on spherical harmonics, analyzed and investigated the key technologies in 3D model retrieval.The most important part of feature extraction algorithm based on spherical harmonics was to define spherical function which describes centain feature of 3D models. In this paper, four spherical functions based on 3D model geometry feature was defined, each spherical function describe 3D model from a certain point of view. Then, spherical harmonic transform was used on these functions and the energy of the coefficients of spherical harmonics was used to form four 3D feature descriptors. The experimental results indicate that those four feature descriptors have pretty good retrieval performance and robustness to rotate, simplification, scale and noise.Because the discrimination power of feature descriptors based on single spherical function was limited, two types of extended feature descriptors was proposed: feature descriptors based on multiple spherical function and feature descriptors based on concentric sphere partition. These two types of extended feature descriotor improved feature descriptors based on single spherical function from two aspects and those two methods could be integrated to construct feature descriptors based on multiple spherical function sequence. This paper presented the basic frame of feature extraction algoritm of those types of feature descriptor. The experimental results indicate that those two methods could improve retrieval performance of feature descriptor, and the feature descriptor based on multiple spherical function sequence had the best retrieval performance.