Facial expression recognition using depth map estimation of Light Field Camera

Facial expressions recognition has gained a growing attention from industry and also academics, because it could be widely used in many field such as Human Computer Interface (HCI) and medical assessment. In this paper, we evaluate the strength of the Light Field Camera for facial expression recognition. The light filed camera can capture the directions of the incoming light rays which is not possible with a conventional 2D camera. In addition, the light filed camera could estimates depth maps which provide further information to handle the facial expression recognition problem. Firstly, a new facial expression dataset is collected by the light field camera. The depth map is estimated and applied on Histogram Oriented Gradient (HOG) to encode these facial components as features. Then, a linear SVM is trained to perform the facial expression classification. Performance of the proposed approach is evaluated using the new dataset with estimated depth map. Experimental results show that significant improvements on accuracy are achieved as compared to the traditional approach.