Multiple Instances Based Emotion Detection Using Discriminant Feature Tracking

Automatic recognition of emotions from the facial expressions continues to be an important aspect in the field of evolution of new age computing systems. Emotion detection capable computer systems is an ongoing research field that has a numerous applications ranging from recommender systems, human-computer interaction, robotics and affective systems. The various features that increase the complexity of emotion recognition systems include ethnicity, gender, pose, occlusion, beard, moustache etc. The type of database used for learning by systems is of crucial importance. Many databases exist for this purpose but none of them is for posed Indian faces. In this paper,we bridge this gap by providing Bharat Database which contains facial images of Indian people. The database has posed expressions of 102 participants and has 896 images. The participants were asked to pose for different emotions by showing them images eliciting those emotions as well as with the help of expert artists. The annotation was done using polling by a panel of three experts.Experiments were conducted on the database using different algorithms and results are presented for reference. This database will further help the community involved in developing of algorithms for emotion recognition. In this paper we propose two emotion detection approaches, the first one is based on Compact Local Binary Pattern and is used for construction of hybrid features which increases emotion detection accuracy. Second approach is Enhanced Feature Extraction using multiple Patches face on images from indigenously developed Bharat Database of Indian Faces, Japanese Female Facial Expression Database and Karolinska Directed Emotional Faces. The results show its applicability for construction of emotion detection systems.