Skip to Main Content
Discriminating 3D models through the comparison of their 2D projected images from multiple viewpoints, view-based methods conform more closely to human visual recognition for 3D model retrieval. In this paper, a new view-based method using local visual features is proposed. After the projected images of a 3D CAD model are generated, the local features are extracted and represented by the PHOG descriptor for each image. Thus, the similarity between a query model and the object in database is measured by aggregating the similarity of the projected images from all the corresponding viewpoints. At the same time, in order to tackle the “semantic gap problem”, a Support Vector Machine (SVM) is utilized to predict the class which the query most likely belonging to. Our experimental results show the effectiveness of the proposed method.