Joint probabilistic approach for real-time face recognition with transfer learning

Face recognition is an integral part in robot partner's interaction with its human subject. With the task expected, open universe scenario face recognition needs to be performed under uncontrolled and uncooperative environment in real-time without imposing strain on its low powered processors. With the concept of informationally structured space where additional samples can contribute to constructing the prior for recognition, joint probabilistic face method is used for its accuracy, simplicity and transparency in data handling. A system is built upon the method that enables transfer learning from across different domains and real-time adaptive learning for continual information collection. Test result shows that the approach can perform well compared to other established methods under low number of prototype number. Test result on adaptive learning sheds light to the direction of future design.