<![CDATA[Facial Emotion Recognition (FER) is an important computer vision task because facial expressions provide visual cues for identifying human affective states. However, FER remains challenging in real-world conditions due to illumination variation, head pose, occlusion, image quality, and ambiguous facial expressions. This study aims to develop and evaluate a lightweight desktop-based FER application by integrating YOLOv8n as the face detection component and DeepFace as the emotion classification framework. The application was developed using Python, OpenCV, CustomTkinter, Pillow, and multithreading to support image upload and camera-based processing without freezing the graphical interface. The system focuses on four emotion classes, namely happy, sad, neutral, and angry. The proposed workflow consists of image resizing, face detection, facial region extraction, emotion classification, label filtering, and desktop-based output visualization. Evaluation was conducted in two stages: preliminary application testing using ten selected facial expression images and benchmark testing using a balanced FER-2013 subset consisting of 400 images from four target classes. The preliminary evaluation showed that eight out of ten images were classified consistently with human interpretation, producing an initial application accuracy of 80%. The benchmark evaluation achieved an accuracy of 76.4%, with macro-averaged precision of 0.74, macro-averaged recall of 0.73, and macro-averaged F1-score of 0.73. The results indicate that the YOLOv8n–DeepFace framework is feasible for lightweight desktop-based FER implementation, although ambiguous angry and neutral expressions remain difficult to distinguish. This framework is useful for applied computer vision education, prototype development, and preliminary FER deployment studies.]]>
