Recognition of facial expression based on CNN-CBP features

Automatical recognition of facial expression is an interesting and challenging problem, which has so many applications such as expression synthesis, human-robot interaction, metal state identification, intelligent tutoring systems, operator fatigue, music for mood, and clinical medicine. The vital step of a successful approach is deriving features from raw facial image. The existed methods of features extraction are the hand-crafted features based on geometric features or appearance features, and the auto-learned features. To utilize the benefit of low computation of hand-crafted features and the high-representation of auto-learned features, we firstly proposed the combined features CNN-CBP with putting together Centralized Binary Patterns (CBP) features and Convolutional Neural Network (CNN) features. And then, we classified the features using Support Vector Machine (SVM). With the help of the CNN-CBP features, we achieved average recognition accuracy of 97.6% on the Extended Cohn-Kanade datasets and 88.7% on the Japanese Femal Facial Expression datasets based on 10-cross validation.