PVRAR: Point-View Relation Neural Network Embedded with Both Attention Mechanism and Radon Transform for 3D Shape Recognition

keywords: Multimodal fusion, 2D linear Radon transform, attention mechanism, 3D shape recognition
Owing to the favorable performance of deep neural networks for 3D shape recognition, an increasing number of researchers are interested in designing novel 3D shape descriptors. However, the relationship between multiple views and point clouds needs to be further elucidated. We propose a multimodal method that combines the features of point clouds and multiple views, i.e., point-view relation neural network embedded with both attention mechanism and Radon transform, to obtain better descriptors. First, a two-dimensional linear Radon transform is performed to investigate linear and color features in multiple views, and the features are used as the input of our network to enable significant distinctions between different views. Moreover, a convolutional block attention module is adopted to enhance the features of point clouds and hence improve the expression ability of feature descriptors. The effectiveness of the proposed method is verified using ModelNet40 and ModelNet10 datasets. Experimental results show that our method can effectively improve the capability of feature extraction and expression as well as achieve state-of-the-art performance on two well-known 3D datasets.
mathematics subject classification 2000: 68T10
reference: Vol. 40, 2021, No. 6, pp. 1217–1243