Title:	PROBABILISTIC AND BOTTLENECK TANDEM FEATURES FOR EMOTION RECOGNITION FROM VIDEOS
Author(s):	Salma Kasraoui, Zied Lachiri and Kurosh Madani
ISBN:	978-989-8533-79-1
Editors:	Katherine Blashki and Yingcai Xiao
Year:	2018
Edition:	Single
Keywords:	FER, LBP-TOP, Tandem Features, MLP, SVM
Type:	Short Paper
First Page:	387
Last Page:	391
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	This paper proposes a tandem based approach for handcrafted features optimization in facial expression recognition (FER) in the wild. The approach trains an artificial neural network (ANN) to generate new trainable features: probabilistic features and bottleneck features. The former are generated from the class probabilities estimates and the latter are obtained from the activations of a narrow hidden layer in the ANN. In this work, we train a multi-layer perceptron (MLP) on Local Binary Patterns on Three Orthogonal Planes (LBP-TOP) features to extract the so-called probabilistic and bottleneck features. Support Vector Machines (SVM) are used to predict one of the seven emotion categories (anger, disgust, fear, happiness, sadness, surprise and neutral) from the Acted Facial Expression in the Wild (AFEW) database. Noteworthy improvements are shown in the classification accuracy.
	Go Back