Non-rigid 3D Shape Classification Based on Low-Level Features

Abstract

Non-rigid 3D shape classification is an important issue in digital geometry processing. In this paper, we propose a novel non-rigid 3D shape classification method using Convolutional Neural Networks (CNNs) based on the scale-invariant heat kernel signature (SIHKS). Firstly, SIHKS feature is extracted and we can get a matrix for every 3D shape. Then CNNs is employed to shape classification. The matrix of 3D shapes can be the input of CNNs. Finally, we can obtain the category probability of 3D shapes. Experimental results demonstrate the proposed method can get better results compared with SVM.