A Convolutional Neural Network-Based Approach for Emotion Recognition Using Facial Expressions

Facial emotion recognition (FER) serves as a significant link that connects human-to-computer interaction and concerns affective computing and mental health since it is possible for the machine to understand and react to human feelings in real-time. A deep convolutional neural network was constructed as part of this research to distinguish anger, contempt, disgust, fear, happy, sadness, and surprise which are the seven target feelings present in pictures from the CK + dataset. Specifically, the original and final implementation of the Convolution Neural-Network (CNN) architecture can be described as comprising a sequence of multiple convolutional layers and a sequence of dense layers that are used for high-reliability emotion class predictions. During model training, 48 by 48 pixels were used on the train test validation images. Various forms of dropout regularization were used to reduce overfitting and enhance model robustness. F1 scores, precision and recall metrics along with other evaluation metrics were deployed to assess model performance with all the categories achieving a reasonable classification success ratio. Therefore, the CNN model can be regarded as a quite dependable tool for emotion recognition with also extending to practical FER primary operations.