Evaluation of Local Shape Descriptors for 3D Shape Retrieval

摘要

Collections of 3D models are becoming prominent in many fields. As the number and sizes of such databases grow, so does the need to accurately search for and retrieve 3D models. Thus, a standard 3D shape retrieval problem becomes returning all other objects in the database ordered in decreasing similarity to a given query. To approach this problem, a number of shape descriptors have been proposed. Global descriptors, such as area and volume (1), shape distributions (2), ratios derived from the objects convex-hull (3), or the LightField Descriptor (4), return a feature vector, a single vector of values to represent an object. The distance between two feature vectors then quantitatively represents the dissimilarity between their corresponding objects; the more similar the objects are, the lower their dissimilarity value, and a value of 0 indicates they are identical.