Fig. 3
From: Spectral-spatial feature fusion for real-time facial expression recognition
Effect of integrating spectral-domain information via FFT in the SPA module. The image on the right shows the attention distribution from the original SPA module without frequency modeling. After incorporating the FFT-based spectral extractor (left), the model focuses more precisely on semantically discriminative facial regions such as the mouth and eyes. This highlights the importance of frequency-aware attention in enhancing subtle expression cues.
