Facial Emotion Recognition Method Based on Canny Edge Detection Using Convolutional Neural Network

Facial Emotion Recognition(FER) has made a pivotal contribution to Human Society by accomplishing complex tasks which humans struggle to achieve. The traditional Machine Learning (ML) models that follow explicit feature extraction from images require domain expertise, manual effort and also lack spatial invariance. The primary cause for the hindrance of the recognition accuracy is the imbalance of classes that exist in the datasets along with the limited availability of training data. To address these limitations, a facial emotion recognition method based on Canny edge detection utilizing a convolutional neural network is proposed in this paper. Important structural information from facial images is extracted to intensify the recognition rate, and techniques such as data augmentation, and oversampling are utilized. The proposed Convolutional Neural Network (CNN) model is trained using the dataset FER2013. The accuracy achieved with only 40 iterations of training was 76.92 percent with the devised method that surpasses other CNN architectural methods such as traditional CNN, Xception and AlexNet.