Authors: Pradeepa S, Anamika S S, Mariammal alias Madhu M, Oviya S
Abstract: Facial expressions provide useful visual cues for human–computer interaction, but automatic interpretation remains challenging because expressions vary across individuals, lighting conditions, head pose, and image quality. This paper presents a face emotion recognition system that combines computer vision with a convolutional neural network (CNN) to classify facial expressions from images and live video. The system detects a face using OpenCV’s Haar Cascade face detector, extracts and preprocesses the facial region, and passes a 48 × 48 grayscale image to a CNN trained using the FER-2013 dataset. The network uses convolutional blocks, batch normalization, max pooling, dropout, and a softmax output layer to classify seven expression categories: angry, disgust, fear, happy, neutral, sad, and surprise. The implementation also integrates separate pretrained age-group and gender models, allowing the application to display multiple facial attributes in a unified interface. Emotion confidence scores and bounding boxes are overlaid on the detected face, while temporal averaging across recent frames is used to reduce rapid changes in predictions. The project demonstrates an end-to-end facial analysis pipeline using Python, TensorFlow/Keras, and OpenCV. Quantitative performance should be reported from the final held-out test run; the project report supplied for this paper does not provide a verified final test-results table, so no unverified performance claim is made here.
DOI: https://doi.org/10.5281/zenodo.23417701
