Automatic Analysis of partially occluded Facial Expressions

Abstract

The most important factor that degrades the performance of a facial expression recognition system is the presence of face occlusions due to scarves, hat, sunglasses, mask etc., Occlusion results in loss of discriminative information, particularly with the case of lower face occlusion, the mouth region where most of the emotions are expressed. Hence a novel block based approach to deal with expression recognition in the presence of partial occlusion has been examined to confirm the portion of the face that holds the foremost discriminative part for emotion classification. Four different types of occlusions namely- eye or upper, mouth or lower, right and left face occlusions are handled in this paper and bayesian network (BN) and support vector machine (SVM) are exploited in emotion recognition from the occluded facial expressions. Experiments are conducted with images of varying block sizes.