Enhancing low-light images through HVI color space with spatial-frequency domain integration
Low-light image enhancement remains challenging because brightness amplification often introduces color bias, chromatic noise, and loss of fine details. To address these issues, we propose SFH-Net, an horizontal-vertical-intensity (HVI)-guided luminance-chrominance collaborative enhancement framework. The proposed method operates in the HVI color space, where the H/V chrominance channels and the intensity channel are processed through dedicated branches. In the luminance branch, a frequency-enhanced residual block with a fixed center-square frequency partition provides spectral auxiliary cues for illumination and texture restoration, followed by spatial residual refinement. In the chrominance branch, a U-Net-based chrominance denoiser module predicts signed residual corrections for the H/V channels, suppressing chromatic noise while preserving hue-direction information. During training, a two-stage strategy first stabilizes reconstruction and then introduces adversarial chrominance refinement. Experiments demonstrate that SFH-Net achieves a better trade-off among reconstruction accuracy, structural fidelity, and parameter compactness. The source code is available at: https://github.com/Zhanghuijie-one/SFH-Net.