Abstract

3D object recognition is a longstanding task in computer vision and has shown wide applications in computer aided design, virtual reality, etc. Current state-of-the-art methods mainly focus on 3D object representation for recognition. Concerning the multi-modal representations in practice, how to effectively combine such multi-modal information for recognition is still a challenging and urgent requirement. In this paper, we aim to conduct 3D object recognition using multi-modal information through a cross di usion process on multi-hypergraph structure. Given multi-modal representations of 3D objects, the correlation among these objects is formulated using the multi-hypergraph structure each representation separately, which is able to model complex relationship among objects. To combine multi-modal representation, we propose a cross discussion process on multi-hypergraph, in which the label information is propagated from multiple hypergraphs alternatively. In this way, the multi-modal information can be jointly combined through this cross di usion process in multi-hypergraph structure.We have applied the proposed method in 3D object recognition using multiple representations. To evaluate the performance of the proposed cross di usion method, we provide extensive experiments on two public 3D object datasets. Experimental results demonstrate that the proposed method can achieve satisfied multi-modal combination performance and outperform the current state-of-the-art methods.

Download paper here