FSCT-UIE: Frequency-Spatial Collaborative Transformer for Underwater Image Enhancement

Underwater images often suffer from degradation due to light absorption and scattering, causing low contrast, color distortion, and blurred details. Existing enhancement methods struggle to address both global and local degradations effectively. This paper proposes FSCT-UIE (Frequency-Spatial Collaborative Transformer for Underwater Image Enhancement), integrating high-frequency directional modeling, low-frequency structural modeling, and frequency-spatial fusion. The framework uses directional subband attention, frequency-channel coupled convolution, and dual-branch fusion to enhance global color consistency and local details. Experiments on LSUI and UIED datasets show FSCT-UIE achieves PSNR scores of 28.29 dB and 25.12 dB (up to 4.39 dB higher than existing methods), SSIM scores of 0.938 and 0.926 (maximum improvement of 0.117), and UIQM scores of 4.452 and 3.891 (maximum improvement of 0.218), validating its effectiveness in global color correction and local detail enhancement.