3D shape classification and retrieval based on polar view.

Abstract This paper presents a novel approach for classifying and retrieving 3D shapes. We first propose a new polar view based strategy that, by using 360-degree projection, effectively reflects the internal structure and key features of a 3D shape. Specifically, the point cloud of the 3D shape is mapped to a two-dimensional (2D) plane, and the polar view representation is obtained by the maximum depth of the point cloud. Projecting 3D point clouds into a 2D plane enables us to use sufficient image data for training, and the process also takes advantage of both image-based and 3D shape-based methods. We propose a convolutional neural network (CNN) that is trained by using only a single polar view of a 3D shape to obtain the polar view representation (PVR) and realize the classification and retrieval of 3D shapes. Experiments on standard datasets, such as ModelNet10 and ModelNet40, show that our method outperforms state-of-the-art methods on 3D shape classification and retrieval. [ABSTRACT FROM AUTHOR]