DWF-Net: A Wavelet-Based Dual-Frequency Fusion Network for Image Blood-Stain Removal
Wenfeng Huang, Ruizhong Wang, Xiangyun Liao, Yinling Qian, Qiong Wang · 2025
In endoscopic surgeries, the camera lens is frequently contaminated by blood, which significantly deteriorates image quality and impairs accurate clinical observation and diagnosis. We propose a lightweight multi-scale wavelet-based dualfrequency fusion network(DWF-Net) tailored for efficient bloodstain removal. Our model integrates multi-scale feature extraction and wavelet-based decomposition, enabling it to capture both global context and fine-grained local details. By operating in the wavelet domain, the network effectively separates and suppresses blood-stained regions through low- and high-frequency coefficient fusion, while the inverse transform recovers critical image details. Compared to SOTAs, our design emphasizes efficiency, featuring fewer parameters(only 460 K). Experimental results show that our method outperforms SOTAs in both blood-stain removal quality and parameter efficiency, highlighting its strong potential for practical clinical applications.