Facial Emotion Recognition (FER) has emerged as a significant research area in artificial intelligence and affective computing due to its wide range of real-world applications. Human emotions expressed through facial features provide valuable insights into psychological and behavioural states. This paper presents a lightweight Convolutional Neural Network (CNN)-based facial emotion recognition system designed for real-time applications. The proposed model is trained using the FER-2013 dataset, which consists of grayscale facial images categorized into seven emotional classes: Angry, Disgust, Fear, Happy, Neutral, Sad, and Surprise. Unlike deep and computationally intensive architectures, this work focuses on developing a shallow and efficient CNN model that can operate on standard hardware without requiring GPUs. Haar Cascade classifiers are employed for real-time face detection, and OpenCV is integrated to capture and process live webcam input. Each detected face is preprocessed and resized to meet the input requirements of the CNN model before emotion classification. The system demonstrates reliable performance under normal lighting conditions while maintaining low computational cost. Experimental observations indicate satisfactory accuracy with smooth real-time execution. The proposed solution is beginner-friendly, easily deployable, and suitable for academic and low-resource environments. This work proves that effective facial emotion recognition can be achieved using optimized and accessible deep learning techniques without complex infrastructure.

Keywords: Facial Emotion Recognition, Convolutional Neural Network, Deep Learning, FER-2013 Dataset, OpenCV, Real-Time Detection