Avaliação quantitativa de desempenho do registro de imagens tridimensionais de faces por meio do algoritmo CP-ICP

Abstract

In 3D face reconstruction applications, an important issue is the matching between di erent samples of a given face.This is a point cloud registration task and is traditionally solved using the Iterative Closest Point (ICP) algorithm. In order to improve the e ciency of this technique, in this article we propose a preprocessing step to the ICP algorithm. We check the quality of the registration using as metrics the RMSE (Root Mean Square Error) measure, measures derived from the rotation matrix output by ICP, as well as measures based on both the SSIM (Structural Similarity Index) and entropy . Our experiments were carried out on a set of 3D data of faces and considered both rigid and non-rigid registration scenarios.