Face Analysis Under Uncontrolled Conditions

6Facial Motion Characteristics

Benjamin ALLAERT¹, Ioan Marius BILASCO² and Chaabane DJERABA²

¹IMT Nord Europe, Lille, France

²University of Lille, France

6.1. Introduction

Descriptors based on dense motion characterization (i.e. computation of the motion in each pixel of the image) have proven their efficiency in facial expression analysis, and seem to be better adapted to characterize the dynamics of facial expressions. Although many facial expression analysis processes have been proposed in the literature, it is difficult to find a process as well adapted to characterize low- and high-intensity facial movements. This difficulty is directly related to the motion characteristics of these expressions. In the presence of macro-expressions, facial deformations induced by facial muscles are easily perceptible. However, in the presence of micro-expressions, where the movement intensities are very low, special attention must be paid to encode the subtle deformations.

During the acquisition of faces, the appearance of noise (i.e. motion discontinuities) coming from different factors (illumination, sensor noise, occlusions) reinforces the difficulty of the analysis of facial movements. In addition to the acquisition noise, the analysis of the face is delicate because some facial deformations cause the appearance or disappearance of wrinkles, which cause motion discontinuities. This requires adapting the facial motion characterization process to reinforce the distinction between noise ...