Facial Expression Recognition Utilizing Multi-scale Energy Regional Attributes and Spiking Attention Mechanisms

This study presents a facial expression recognition algorithm that integrates multiscale Ising models and energy region attribute feature extraction, combined with spiking neural networks and attention mechanisms. The algorithm, based on the thermal energy attributes of different facial regions, generates multiscale features by adjusting temperature parameters to effectively capture multidimensional features. The study suggests that different attribute regions of the face have varying degrees of influence on expression recognition, reflected in thermal maps. By extracting multiscale features through spin states of the Ising model, this method reduces computational complexity and improves recognition accuracy. Experimental validation on the AffectNet dataset demonstrates a 10% improvement in recognition accuracy.