Multichannel CNN for Facial Expression Recognition

In the past years there have been several attempts on the task of facial expression recognition. We have developed a new method based on the understanding of CNN and various image processing techniques. A multi-channel CNN architecture is proposed, which helps in performing improved facial expression recognition on frontal face images. For better feature extraction, fine tuning of images has been done by different preprocessing methods, namely Sobel edge detection, median filtering and Gaussian smoothing. Thereafter, the preprocessed images, have been fed in a novel manner in the proposed multi-channel CNN model. The model is evaluated on three challenging benchmark datasets - JAFFE, CK+ and Oulu-CASIA. The performance is comparable with various state-of-the-art approaches for facial expression recognition, which is evident from the results obtained.