Facial expression recognition using feature level fusion

Abstract

In this paper, a model for facial expression recognition (FER) fusing local and global features is proposed. Local features are extracted by dividing the face into multiple regions and carefully selecting regions of interest which assists in reducing redundancy. Global features are extracted from the entire subject face and features of interest are extracted for expression recognition. The major drawback with previously available techniques in literature is that they do not take finer details into account along with global geometric features. In our work, using concatenation fusion, FER is performed in an effective way; one part of model recognizes facial expressions on a complete scale while the other performs the same task on a finer scale, thereby improving the accuracy. Accuracy with the proposed fused model is increased to 93.52% in comparison to the accuracy of 74.11% with local binary patterns and 90.64% with global features when used as standalone technique.