A Hybrid 3D Descriptor With Global Structural Frames and Local Signatures of Histograms

An effective feature descriptor is important for the 3-D object recognition and reconstruction. This paper presents a novel 3-D Frame-signature of histograms of orientation (SHOT) descriptor which combines the global structural frame with local surface information. Global feature descriptors are more descriptive for surface representation, while sensitive to occlusion. In contrast, the local feature descriptors are more robust in scenes with occlusion, while less discriminative due to the limit scope of the local neighborhood. The proposed hybrid Frame-SHOT descriptor uses the SHOT feature to represent the local surface and uses the structural frame points to encode the global surface. Both local and global information for surface description are combined by the hybrid 3-D descriptor. We have compared the proposed Frame-SHOT descriptor with the existing global and local descriptors on four public data sets for descriptor matching, 3-D object recognition, and point cloud registration. The experimental results validate the effectiveness and the robustness of the proposed descriptor.