Relevance analysis of 3D shape descriptors on interest points and regions of the face

Abstract

The human face is made from different morphological structures that can be represented in 3D by descriptors. This work presents an analysis of the relevance of descriptors such as main curvatures, mean curvature, Gaussian curvature, shape index, curvedness, multi-resolution spin images, and the multi-scale shape index, all of them in regions of the human face. This study aims to determine which descriptors offer the most relevant information in particular locations of the face by using the Fisher’s discriminant analysis. Therefore, the most adequate descriptor that can be used in facial anthropometry or face recognition can be identified.