Facial Expression Recognition Using a Hybrid CNN-SIFT Aggregator

Recognizing facial expression has remained a challenging task in computer vision. Deriving an effective facial expression recognition is an important step for successful human-computer interaction systems. This paper describes a novel approach towards facial expression recognition task. It is motivated by the success of Convolutional Neural Networks (CNN) on face recognition problems. Unlike other works, we focus on getting good accuracy results while requiring only a small sample data to train the model by merging the CNN and SIFT features. The proposed classification model is an aggregation of multiple deep convolutional neural networks and a hybrid CNN-SIFT classifiers. The goal of using SIFT features is to increase the performance on small data as SIFT does not require large training data to generate useful features. The model has been tested on FER-2013, SFEW 2.0 and CK+. The results showed how CNN-SIFT features improve the accuracy when added as a voting member in an ensemble classifier. It generates state-of-art results on FER-2013 and CK+ datasets, where it achieved 73.58% on FER-2013 and 99.35% on CK+.