Emotion and Advertising Effectiveness: A Novel Facial Expression Analysis Approach Using Federated Learning

Facial expression recognition using computer vision systems is a valuable technology with diverse applications. However, traditional 2D image analysis techniques face challenges such as varying illumination conditions and head orientation, making accurate recognition difficult. Deep learning-based models require large annotated datasets, raising privacy concerns and limiting access to real-world facial expression data. To address these issues, this research proposes a decentralized facial expression recognition model using federated learning, a privacy-preserving deep learning paradigm. The research also incorporates a graphical user interface (GUI) to facilitate facial image input during advertisement playback for emotion recognition. In this research FER-2013 and CK+ datasets are utilized, which encompass seven different facial emotions. Three robust deep learning architectures (CNN, ResNet-50, and VGG-16) are employed for classifying facial expressions. The proposed models are evaluated against state-of-the-art emotion classification models to assess their performance. The results demonstrate that the CNN model trained using federated learning achieves a maximum accuracy of 77%, with a test accuracy of 72%. Additionally, the GUI allows users to conveniently input facial images while advertisements are playing, enabling real-time emotion recognition. By leveraging federated learning, this research addresses privacy concerns and facilitates decentralized facial expression recognition. The GUI component enhances user experience, providing a user-friendly interface for capturing facial images during advertisement playback, thereby enabling effective emotion recognition in real-time.