Discriminative Regions Selection for Facial Expression Recognition

Human Machine Interaction systems are able to perceive facial expressions more naturally and reliably. In this paper, we introduced a new idea to recognize facial expression by selecting the most discriminative facial regions relying on facial expression appearance. The proposed approach is based on the prior knowledge of psychology studies which show that only some facial regions are descriptive in expression revelation. In fact, regions selection seeks to collect the descriptive regions which are responsible of expression divulgence and this was performed using Mutual Information technique. Regarding facial feature extraction, we applied Local Binary Pattern technique to encode facial expression micro-patterns. An experimental study shows that using descriptive regions improved facial expression classification accuracy as well as reduced features vector size. Indeed, we attested the independency of the selected regions of the dataset and the descriptors.