Facial emotion detection considering partial occlusion of face using Bayesian network

Recently, robots that communicate with human have attracted much attention in the research field of robotics. In communication between human, almost all human recognize the subtleties of emotion in each other's facial expressions, voices, and motions. Robots can communicate more smoothly with human as they detect human emotions and respond with appropriate behaviors. Usually, almost all human express their own emotions with their facial expressions. In this paper, we propose an emotion detection system with facial features using a Bayesian network. In actual communication, it is possible that some parts of the face will be occluded by adornments such as glasses or a hat. In previous studies on facial recognition, these studies have been had the process to fill in the gaps of occluded features after capturing facial features from each image. However, not all occluded features can always be filled in the gaps accurately. Therefore, it is difficult for robots to detect emotions accurately in real-time communication. For this reason, we propose an emotion detection system taking into consideration partial occlusion of the face using causal relations between facial features. Bayesian network classifiers infer from the dependencies among the target attribute and explanatory variables. This characteristic of Bayesian network makes our proposed system can detect emotions without filling in the gaps of occluded features. In the experiments, the proposed system succeeded in detecting emotions with high recognition rates even though some facial features were occluded.