Analysis and Recognition of Facial Expression Based on Point-Wise Motion Energy

Automatic estimation of facial expression is an important step in enhancing the capability of human-machine interfaces. In this research, we present a novel method that analyses and recognizes facial expression based on point-wise motion energy. The proposed method is simple because we exploit a few motion energy values, which is acquired by an intensity-based thresholding and counting algorithm. The method consists of two steps: analysis and recognition. At the analysis step, we compute the motion energies of facial features and compare them with each other to figure out the normative properties of each expression. We extract the dominant facial features related to each expression among facial features. At the recognition step, we perform rule-based facial expression recognition on arbitrary images using the results of analysis. We apply the proposed method to the JAFFE database and verify its feasibility. In addition, we implement a real-time system that recognizes facial expression very well under weakly-controlled environments.