The effect of spatial information characterization on 3D local feature descriptors: A quantitative evaluation

Designing local feature descriptors for 3D objects is a fundamental yet challenging task in 3D computer vision. Both geometry and spatial information descriptions are critical for a 3D local descriptor, while most previous studies concentrate on the former one. This paper investigates on how the characterization of spatial information would affect a 3D local descriptor in terms of descriptiveness, robustness, compactness and efficiency. The evaluation process is deployed as follows. First, based on the analysis of representative spatial information characterization methods of existing local shape descriptors, six typical characterization methods with different spatial dimensions and partition principles of spatial information are presented. Second, three geometric attributes, i.e., normal deviation, local depth and shape index, are respectively assigned to each point in the local surface for local geometry description, creating a total of 18 different feature descriptors. Then, a quantitative analysis of performance (i.e., descriptiveness, robustness, compactness and efficiency) for these descriptors is carried out on three benchmark datasets. Grounded on the experimental outcomes, the traits, merits and demerits of each spatial information encoding approach are eventually summarized. This study reveals that different spatial information encoding approaches would bring significant effect on a local shape descriptor with respect to its discriminative power, stability, compactness and efficiency.