三维模型多示例生成算法的应用.docVIP

  The Applications of 3D Models Multi-instances Generation Algorithm# Wan Lili* 5 10 15 20 25 30 35 40 (Institute of Information Science, Beijing Jiaotong University, Beijing 100044) Abstract: In shape matching, retrieval and classification, a 3D model is always taken as a whole to do feature extraction and similarity measure, but in fact many 3D models from the internet are composed of multiple 3D objects. 3D models multi-instances generation algorithm can solve this problem, whose key idea is to divide 3D model consisted of several objects into some instances. In this paper, we discuss two applications which can highly benefit from this alogrithm. These applications include 3D model retrieval and semantic annotation. In the first application, the key is the shape similarity between the query model and the model having multiple objects. In the second application, the semantic annotation of each instance is preferentially solved. Experiments show the two applications achieve good results. Keywords: 3D model retrieval; 3D Model semantic annotation; shape analysis 0 Introduction 3D models have been widely used in mechanical engineering, digital entertainment, biological medicine, computer education, and a variety of other fields. As the number of 3D models online and offline has increased dramatically over the last few decades, there is an increasing need for acquiring 3D models quickly and conveniently. Early researches focus on extracting shape features from the models and measuring shape similarities, and some 3D model search engines have been proposed [1~3]. The most common search interface is to use keywords, but 3D models lack semantic descriptions from themselves, so shape classification and automatic annotation approaches have been investigated. Huber et al. [4] extract parts from training models and group into part classes, and then derive a mapping from part classes to object classes, finally utilize Bayesian decision for classifying vehicles into a set