3D Face Expression Recognition via Geometry A Preparatory Path

Abstract

This work presents a preliminary automatic 3D facial expression recognition procedure based on point-by-point mapping of seventeen Differential Geometry descriptors onto the query facial depth map, which is then partitioned into seventy-nine regions. Then, mean, median, mode, volumes, histograms are computed for each region and for each descriptor, and adopted as features, to reach a diverse set of parameters representing the shape description of the query face. Each parameter set, given by a geometrical descriptor, a region, and a feature, form a trio, whose featuring numerical values are compared with appropriate thresholds, set via experimentation. This allows the identification of the emotion-based expression of the probe 3D face among the six basic ones (anger, disgust, fear, joy, sadness, surprise). The algorithm has been tested on the Bosphorus database and is conceived to be applied to security, marketing, medical contexts. The three-dimensional context has been chosen due to its invariance to different lightening/make-up/camouflage conditions.