A partial shape matching technique for 3D model retrieval systems

Abstract

The use of 3D models is gaining wide popularity since they are very important for computer graphics applications. Recently, similarity search techniques for 3D models have been investigated intensively to retrieve 3D models from the Internet. The techniques extract shape descriptors from 3D models and use these descriptors for indices of databases. Since the shape descriptors can be extracted using software, it is more efficient compared with keyword-based indices. Various shape descriptors have been proposed to improve shape similarly search results in relation to invariance of rotation, scale and translation. Most similarity search techniques are suitable for comparing each individual 3D model from databases. However, our similarity search techniques can compare not only each individual 3D model, but also similar portions of 3D models. Using our technique, each 3D model is divided into a huge number of parts, and shape descriptors are extracted from these parts to compare similarities. Although there are a large number of combinations for comparing the similarities for portions of 3D models, our shape descriptor extraction technique enables fast computing for evaluating similarities.Our system automatically decomposes 3D models into several parts by comparing angles created by normal vectors of each polygonal face. The system finds narrow angles and cuts polygonal faces into parts. Once the 3D models are decomposed, the system extracts rotation invariant shape descriptors from each part, and the descriptors are compared using a histogram to evaluate similarities. In our preliminary experiments, about 1700 3D models are decomposed into about 56,000 parts. Although there is a large number of comparisons, our shape descriptor based on equivalence class makes it possible for the system to retrieve 3D models in less than average of 10 seconds. The system allows users to find similar portions of 3D models as shown in Figure 2. Also, the system can list 3D models that contain similar parts as shown in Figure 3. Our experimental search engine can be accessed from the following web site: http://www.nime.ac.jp/web3d/