Personalized broad learning system for facial expression

Understanding human emotions through facial expressions is key enabling technology for interactive robots. Most approaches of facial expression recognition are designed for the average user. It is difficult for them to maintain high accuracy for special users with different cultural backgrounds and personalities, which limits their application in real world scenarios. Personalized classifier is a feasible solution, but it needs to be retrained for new users outside the training set. In this paper we present a framework for personalizing facial expression recognition which does not require re-training models after entering new data. Personalized incremental updating mechanism is achieved by designing a novel broad learning system. Specifically, we propose a transfer learning model based on emotional information entropy as the mapping feature layer to ensure the accuracy of mapping under the condition of small sample size. Then, the weights of our proposed model can be updated by multi-layer singular value decomposition method if incremental data is entered. We exhibit the superiority of our approach in multiple facial expression datasets. Experimental results show that our method has higher accuracy and generalization ability with previous personalization techniques.