SFFNet: Multi-Scale Sparse Focus Feature Cue-Aware for Efficient Facial Expression Recognition

Facial expression recognition is an important subtask in the field of computer vision. Accurate recognition of facial expressions plays a critical role in applications such as human-computer interaction, emotion analysis, and intelligent surveillance. In this work, we propose a novel approach for learning multi-scale sparse focus features by combining CNN and Swin Transformer architectures, which enables the effective capture of both local and global patterns in facial expression images. Additionally, we introduce two key modules: Adaptive Hierarchical Feature Construction (AHFC) and Dual Contextual Attention Synthesis Unit (DCASU). Specifically, the AHFC module leverages Swin Transformer to extract multi-scale facial expression features while mining the relationships among sparse focus features. The DCASU module, on the other hand, extracts local features and utilizes attention mechanisms to focus on crucial regions of the face. Extensive experiments demonstrate that our method achieves state-of-the-art performance on the RAF-DB (89.18%) and KDEF (95.81%) datasets, validating the effectiveness and superiority of our approach.