Facial expression recognition (FER) plays a critical role in affective computing and human-computer interaction. However, in low-light conditions, traditional FER models suffer from reduced accuracy due to poor visibility and the loss of fine-grained facial details. To address these limitations, we propose HiFeFace (High-Frequency Feature Fusion for Face), a novel dual-branch FER framework that jointly leverages raw RGB information and high-frequency contour features for robust expression recognition. Specifically, we introduce a learnable High-Frequency Edge (HiFE) module to extract expression-relevant structural cues, and a Multi-Feature Dynamic Fusion (MFDF) module that adaptively integrates multimodal features using spatial and channel attention. Our architecture effectively preserves subtle facial features even under degraded lighting conditions. Extensive experiments on FERPlus, RAF-DB, and its synthetic low-light variants demonstrate the superior performance of HiFeFace, achieving 91.50% accuracy on RAF-DB and 91.32% on FERPlus, outperforming existing state-of-the-art methods. These results highlight the potential of high-frequency contour information in enhancing FER under challenging visual conditions.