Researches On Scale-invariant Non-rigid3D Model Retrieval

In recent years, due to the3D model widely used in many industries, such as design and entertainment, the resource of3D model is growing, and the demand for3D model is also increasing.3D model Retrieval is designed to help users find the model resource quickly and easily. Current research on the3D model retrieval model is mainly for the rigid models, however, these methods are difficult to apply in the presence of abundant non-rigid model with complex change. In this paper, based on the geometric thermonuclear diffusion analysis, we studied the non-rigid3D model retrieval method, and implemented the scale-invariant non-rigid3D model algorithm. We also proposed the non-rigid3D model retrieval method based on compression perception. The main contents of this paper are as follows:1The first part summarizes the existing non-rigid3D model retrieval method, especially for analyzes of non-rigid3D model retrieval method based on the diffusion geometry.2The HKS (Heat Kernel Signature), as the apex descriptor of the model applied into a retrieval method, is an important breakthrough in non-rigid shape retrieval, but the impact of HKS greatly affected by the scale of the model. So in this paper, we use mathematical transformation such as discrete Fourier transform to get SIHKS (Scale invariant Kernel Signature) based on the HKS, to eliminate the influence of scale factor. The SIHKS has the quality of isometric invariance, but also has scale invariance. After extracting the SIHKS of models, each vertex using BOW (Bag of Words) method to change SIHKS into the global shape descriptor, the experiments showed that retrieval accuracy in SIHKS methods better than the HKS methods.3. Aiming at change local shape descriptor into the global shape descriptor, this paper obtained by means of training the local shape descriptor of the dictionary, and then, for each3D model, based on the local shape descriptor of compressed sensing sparse coding, we make all the local shape descriptor of its vertices sparse coding become the global shape descriptor. Experimental results show that it has greatly improved accuracy, compared with the BOW methods.