Segmentação de Regiões de Interesse em Imagens 3D de Faces para Reconhecimento de Indivíduos: uma Abordagem Baseada em Unidades de Ação Muscular e Atributos Geométricos de Covariância

Abstract

In this work is proposed an investigation into different ways on slicing process applied on 3D Point Cloud data, representing individuals faces, on individual face recognition problem. The objective is to find subclouds, called slices, that contain greater significance for the extraction of features aiming at individual recognition applications. Based on the analysis of regions of greater variability of the face and solutions proposed by previous works in this same domain, a comparison is made between the classification results with and without slicing the faces, using geometric characteristics in the information extraction process. The purpose of the comparisons is to investigate if selecting specific regions of the 3D face can improve classification results, two types of segmentation were evaluated: triaxial slicing and pizza slicing with superposition, the latter being a contribution of this project. The results are promising, indicating that a refinement of the technique can generate a classification of individuals robust to the natural deformations of the surface of the face.