Active appearance model applied to emotion recognition through partially occluded facial expression

Abstract

Facial expressions are an important demonstration of human's humours and emotions. Algorithms capable of recognizing facial expressions and associate them to emotions become useful for many applications and the Machine Learning approach shows to be efficient in solving this challenge, which becomes even more difficult when facial regions are occluded. An Active Appearance Model must be trained in order to reconstruct the occluded facial region - whose facial expression and related emotion were previously labeled in static images. Feature selection filters will be applied to these images and dimensionality reduction methods will be used to compress the data. This research project focuses on studying and implementing a robust Active Appearance Model to reconstruct occluded regions of the image and then verify its efficiency in a classifier trained with non-occluded images. Experimental results will be tested on various public databases to evaluate the method performance.