HSV-DehazeNet: hue consistency calibration and haze density supervision for image dehazing

Single image dehazing is a key low-level vision task for enhancing visibility and supporting high-level perception. However, RGB-based methods often suffer from color distortion and residual haze due to strong inter-channel coupling. From the HSV perspective, we observe that the hue-channel distribution changes only slightly before and after dehazing. Accordingly, we design a lightweight calibration strategy that enforces hue consistency to mitigate color shifts. In addition, we exploit the saturation–value discrepancy as a haze-density cue and incorporate it as an auxiliary supervision signal during training. Therefore, we propose HSV-DehazeNet, which incorporates the Hue Consistency Calibration Module (HCCM) and imposes Haze Density Supervision (HDS) to enable color-stable restoration and residual haze suppression. Experiments demonstrate that our method achieves competitive performance on real-world paired datasets and generalizes effectively to unpaired benchmarks. The code is available at https://github.com/littlezuicat/HSV-DehazeNet-master.