Active-contour based-on face emotion patterns

Complex Human-Robot Interactions (HRI), e.g. occurring in assistive-aid applications, require the automatic understanding of human facial expressions in order the intelligent agent could adequately and timely react when facing a particular emotion of its human interlocutor. In this work, we present a new approach to automatically detect and recognize such facial emotions. On one hand, active contours are first applied to detect the face and the changes/moves of its parts (i.e of the mouth, the eyes, etc.), leading to visual patterns which consist in a new quantification of face expressions. On the other hand, an innovative tree-structure based on these detected patterns is proposed in order to effectively recognize the human emotions, such as happiness or surprise, by both increasing the automated process robustness and its computational efficiency. We have evaluated our algorithms on standard face datasets. Our proposed approach has shown excellent performance, outperforming state-of-the-art methods.