Recognition of Emotions through Facial Geometry with Normalized Landmarks

Emotion recognition holds pivotal significance in human social interactions, as it entails the discernment of facial patterns intricately linked to diverse emotional states. The scientific, artistic, medical, and marketing domains have all demonstrated substantial interest in comprehending emotions, resulting in the emergence and refinement of techniques and computational methodologies to facilitate automated emotion recognition. In this study, we introduce a novel method named REGL (Recognizing Emotions through Facial Expression and Landmark normalization) aimed at recognizing facial expressions and human emotions depicted in images. REGL comprises a sequential set of steps designed to minimize sample variability, thereby facilitating a finer calibration of the informative aspects that delineate facial patterns. REGL carries out the normalization of facial fiducial points, called landmarks. Through the use of landmark positions, the reliability of the emotion recognition process is significantly improved. REGL also exploits classifiers explicitly tailored for the accurate identification of facial emotions. As related works, the outcomes of our experimentation yielded an average accuracy over 90% by employing Machine Learning algorithms. Differently, we have experimented REGL with varied architectures and datasets including racial factors. We surpass related works considering the following contributions: the REGL method represents an enhanced approach in terms of hit rate and response time, and REGL generates resilient outcomes by demonstrating reduced reliance on both the training set and classifier architecture. Moreover, REGL demonstrated excellent performance in terms of response time enabling low-cost and real-time processing, particularly suitable for devices with limited processing capabilities, such as cellphones. We intend to foster the advancement of robust assistive technologies, facilitate enhancements in computational synthesis techniques, and computational resources.