Enhancing Emotion Detection Using Metric Learning Approach

Other than speech and body language, facial expression is one of the most prominent ways by which humans communicate their feelings to other humans. The task of detection of emotions in real time accurately has been a very arduous task as methods giving good results are generally computationally exhaustive, whereas the methods that has low computation time do not yield good results. In our system, we have maintained the highest accuracy possible while keeping the computational cost minimal. Various descriptors were tried to test the best trade-off between accuracy and time. We used Distance Metric Learning (DML) for significantly improving the results when the features were mapped to higher dimension. The dataset used is the extended Cohn–Kanade dataset. The system was also tested on subjects that were not available in the dataset and gives a comparable result with other real-time emotion detection system.