Micro-Expression Recognition Based on Multi-Scale Feature Enhancement and Global-Local Feature Analysis

Micro-expression recognition (MER) plays a crucial role in human emotion analysis. However, it remains challenging due to the transient and subtle nature of micro-expressions. To address these issues, a multi-scale feature micro-expression recognition network that integrates a multi-scale feature enhancement module based on an improved Feature Pyramid Network and a global-local feature analysis module is proposed to improve feature representation. The feature pyramid network is deeply optimized for the task of micro-expression recognition. By integrating attention mechanisms and bilinear interpolation, the approach effectively leverages hierarchical feature extraction capabilities to address the spatial misalignment and feature loss issues of feature pyramid network, thereby enhancing recognition performance. Experimental results on the CASME Ⅱ and SAMM datasets demonstrate that the method achieves competitive performance compared to state-of-the-art approaches, particularly excelling in certain key metrics. The approach excelled in the 5-class recognition task, achieving the best results in both unweighted average recall and unweighted F1-score. These findings highlight the effectiveness of the proposed framework in addressing the challenges of MER.