Facial Expression Recognition based on Feelings-Net

摘要

Based on traditional machine learning methods, and the shallow depth study method for facial expression recognition rate low, was proposed based on Feelings - Net network architecture and joined the principal component extraction method of facial expression recognition system, the six basic facial expression recognition in static or video. Viola-jones face recognition framework is adopted to realize the face location and cutting of the input image. Principal component extraction method is used to extract the important parts of facial expressions, which are input into the 12-layer network structure-challenge-net for training and testing. Through experiments in two open source databases (CK+ and JAFFE), the recognition rate was 92.5% and 95.6% respectively, and the frame rate of video facial expression recognition reached 23.4 FPS. Experimental results show that this method is superior to other methods for facial expression recognition.