An Accurate Method for 3D Model Classification using the Explainable Boosting Classifier

Abstract

Accurate and efficient 3D model classification is still a major challenge in computer vision and other fields. This paper presents a new method which is expected to achieve high classification performance. The proposed approach is organised in two major phases. First, we extract a discriminative feature vector by computing three powerful geometric measures, i.e., Shape Diameter Function (SDF), dihedral angles, and the Shape Index, and fuse them into a unified representation for each 3D object. Second, this composite feature vector is used to train an Explainable Boosting Classifier (EBC), a powerful ensemble learning model, to perform the classification task. The effectiveness of the proposed method was thoroughly tested on the Princeton Benchmark dataset. Experimental results show that our proposed method can get a better classification accuracy than several leading state-of-the-art methods.