Genuine vs Posed Expressions: A Juxtaposing of Man and Machine Performance

In face expressions, the genuine vs. posed problem refers to the ability of a computer-based system to analyze an image of a face and to recognize whether facial behavior is due to a real emotion or is the consequence of the voluntary control. In real applications a problem is the lack of fully annotated databases, as in many situations only sequence level annotations are available, thus enforcing a weakly supervised learning framework. In this paper, first we propose a method of automatic identification of spontaneous behavior based on temporal deep neural networks (3D-CNN). The method is further tested in the case of weakly supervised data. To analyze the quality of the results we follow with a subjective experiment where persons are asked to supply the automatic system behavior.