Non-rigid 3D model retrieval using multi-scale local features

Abstract

The number of available non-rigid 3D models in various areas increases steadily. The local features are more effective than global features for the search of these non-rigid 3D models. Global descriptors fail to consistently compensate for the intra-class variability of non-rigid 3D models. To solve this problem, we propose a non-rigid 3D model retrieval method based on multi-scale local features. Firstly, we extract keypoints at multiple scales automaticlly. Then, the Heat Kernel Signature (HKS) local descriptors are computed for each keypoint. However, the HKS descriptors are sensitive to scale. In order to solve this problem, the HKS descriptors are put into the Bag-of-Features (BOF) framework. In the BOF framework, we use a kind of histogram equalization technique to make our feature descriptor robust to model scaling. Experimental results on two public benchmarks show that our algorithm can achieve satisfactory retrieval performance for the non-rigid 3D models.