Facial Expression Recognition using discrete differential operator and Convolutional Neural Networks

Abstract

The facial expressions display the mood of a person that reflect his/her state of mind. Emotions can be positive or negative. The negative emotions affect the mental health as a result of depression, stress and anxiety. In this study, seven emotions i.e. happy, sad, disgust, surprise, contempt, fear and angry are considered under facial expressions recognition (FER). To this end, a CNN architecture containing two convolutional, two pooling and two dense layers are utilized with some additional features between the two convolutional layers. The output of the first convolutional layer called the feature map is multiplied by the original resized image and then fed to the next convolution layer of CNN to generate the information set based feature map. With this model, an accuracy of 98.63 % is achieved. On the other hand, the use of Prewitt and Sobel operators on the input images to produce the preprocessed images followed by the application of CNN on them leads to the recognition accuracy of 99.31%.