A SPATIOTEMPORAL DEEP LEARNING SOLUTION FOR AUTOMATIC MICRO-EXPRESSIONS RECOGNITION FROM LOCAL FACIAL REGIONS

Mouath Aouayeb (1) , Wassim Hamidouche (1) , Kidiyo Kpalma (1) , Amel Benazza-Benyahia (2)

1 IETR - Institut d'Électronique et des Technologies du numéRique
2 URISA - Unité de Recherche en Imagerie Satellitaire et ses Applications

Humans always try to hide their Macro-Expressions (MaE) to conceal their real emotion, and it is hard to distinguish between true and false emotions even with artificial intelligence. Micro-Expressions (MiEs), on the contrary, are spontaneous and fast, undetectable with the naked eye and thus always inform us of true feelings. Therefore , there is plenty of studies to generate an automatic system of detecting and analyzing these MiEs. In this paper we propose a new solution that relies on a combination of Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) applied on particular regions of the face to extract relevant spatial and temporal features, respectively, for MiEs recognition. The proposed solution achieves high recognition accuracy of 90% precision on a different databases including SMIC, CASME II and SAMM. Moreover, under the conditions of Micro-Expression Grand Challenge (MEGC) 2019, our approach performs better than the state of the art solutions including the ones proposed in the challenge.

Keywords: CNN; Index Terms-Micro-Expression; Regions of In- terest; LSTM

Domains: Computer Science [cs]; Image Processing [eess.IV]

Mouath Aouayeb, Wassim Hamidouche, Kidiyo Kpalma, Amel Benazza-Benyahia. A SPATIOTEMPORAL DEEP LEARNING SOLUTION FOR AUTOMATIC MICRO-EXPRESSIONS RECOGNITION FROM LOCAL FACIAL REGIONS. IEEE INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING, Sep 2019, Pittsburgh, United States. ⟨hal-02334439⟩