Deformed 3D model identification using combined depth image

In this paper, we proposed 2D view-based 3D model identification method using depth images, scale-invariant feature transform(SIFT), random sample consensus(RANSAC) and pose normalization. The existing method uses bag-of-feature method using SIFT algorithm. However, the proposed method is not used. In this paper, we perform pose normalization for 3D model to pre-processing. After pre-processing, we get depth images and the combine depth image as one depth image, then use the SIFT algorithm to build feature DB. In the matching process, we remove outlier features to increase match rate using RANSAC algorithm. In the experiment, we compose the database of 3D models from 16 classes in the SHREC benchmark database. Each of classes includes 4 to 5 non-rigid models. The deformed 3D model is used as query 3D model. The match rate of the proposed method is 87.2%.