A simplified and accurate registration for splat-based fruit scans

Abstract

In this paper we first present an efficient and accurate feature-based pairwise registration algorithm between splat-based fruit scans. A new geometric descriptor - oriented splat is introduced as the matching features of fruit scans; a Rough-Refinement matching correspondence process is proposed based on its two properties of position and normals; and transformation is simplified by eliminating the redundant transformation for remitting the computation complexity. Then the pairwise registration is extended to global registration by applying minimal spanning binary-tree to a hierarchical registration. In experiment, the registration results of apple scans are given and the accuracy and speediness of algorithm are illustrated. © 2012 ICIC International.