Discrimination Between Genuine Versus Fake Emotion Using Long-Short Term Memory With Parametric Bias and Facial Landmarks

Xuan-Phung Huynh, Yong-Guk Kim; Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 3065-3072

Abstract


Discriminating between genuine and fake emotion is a new challenge because it is in contrast to the typical facial expression recognition that aims to classify the emotional state of a given facial stimulus. Fake emotion detection could be useful in telling how good an actor is in the movie or in judging a suspect tells the truth or not. To tackle this issue, we propose a new model by combining a mirror neuron modeling and deep recurrent networks, called long-short term memory (LSTM) with parametric bias (PB), by which features are extracted in the spatial-temporal domain from the facial landmarks, and then boil down to two PB vectors: one for genuine and other for fake one. Additionally, a binary classifier based on a gradient boosting is used to enhance discrimination capability between two PB vectors. The highest score from our system was 66.7 % in accuracy, suggesting that this approach could have a potential for useful applications.

Related Material


[pdf]
[bibtex]
@InProceedings{Huynh_2017_ICCV,
author = {Huynh, Xuan-Phung and Kim, Yong-Guk},
title = {Discrimination Between Genuine Versus Fake Emotion Using Long-Short Term Memory With Parametric Bias and Facial Landmarks},
booktitle = {Proceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops},
month = {Oct},
year = {2017}
}