3D Shape Analysis

5Local Shape Descriptors

5.1 Introduction

Local shape description is an essential component for many 3D computer vision and graphics applications. For example, in 3D shape registration, two partial scans of the same object, but captured from two different viewpoints, can be merged together by matching keypoints on the two scans. In 3D object recognition, we may wish to detect and recognize an object occluded in a cluttered 3D scene by matching 3D exemplar models to patches or parts of the 3D scene. In partial 3D shape retrieval, we may wish to retrieve a list of 3D objects that are similar to a given partial scan or a given part of another 3D model. The global shape descriptors described in Chapter 4 are usually not suitable for such tasks. In fact, a global descriptor abstracts the shape of an entire 3D object into one real‐valued vector. As such, the information they convey is not localized to specific parts of the object. This makes them not suitable for tasks such as registration, recognition under occlusions or in cluttered 3D scenes, and partial 3D shape matching and retrieval.

In this chapter, we focus on local descriptors, which, in contrast to global descriptors, encode the shape of small patches around a set of specific keypoints. Thus, they are more robust to occlusions and clutter and are more suitable for partial shape matching and retrieval. Local descriptors can be also aggregated to form global descriptors with a significantly higher descriptive power than the ...