The Diffusion Duet: Harmonizing Dual Channels with Wavelet Suppression for Image Separation

TL;DR

The paper tackles dual-channel blind image separation under unknown mixing by integrating diffusion models with a Wavelet Suppression Module to address mutual coupling between sources. The Dual-Channel Diffusion Separation Model (DCDSM) performs forward diffusion on mixed inputs and uses two conditional reverse denoising branches, augmented by wavelet-domain suppression to disentangle sources progressively. It achieves state-of-the-art results in rain and snow removal and in complex mixture separation, while maintaining reasonable computational costs. The approach offers a principled framework for interactive, multi-scale BIS with practical relevance to adverse-weather cleanup and image restoration tasks.

Abstract

Blind image separation (BIS) refers to the inverse problem of simultaneously estimating and restoring multiple independent source images from a single observation image under conditions of unknown mixing mode and without prior knowledge of the source images. Traditional methods relying on statistical independence assumptions or CNN/GAN variants struggle to characterize complex feature distributions in real scenes, leading to estimation bias, texture distortion, and artifact residue under strong noise and nonlinear mixing. This paper innovatively introduces diffusion models into dual-channel BIS, proposing an efficient Dual-Channel Diffusion Separation Model (DCDSM). DCDSM leverages diffusion models' powerful generative capability to learn source image feature distributions and reconstruct feature structures effectively. A novel Wavelet Suppression Module (WSM) is designed within the dual-branch reverse denoising process, forming an interactive separation network that enhances detail separation by exploiting the mutual coupling noise characteristic between source images. Extensive experiments on synthetic datasets containing rain/snow and complex mixtures demonstrate that DCDSM achieves state-of-the-art performance: 1) In image restoration tasks, it obtains PSNR/SSIM values of 35.0023 dB/0.9549 and 29.8108 dB/0.9243 for rain and snow removal respectively, outperforming Histoformer and LDRCNet by 1.2570 dB/0.9272 dB (PSNR) and 0.0262/0.0289 (SSIM) on average; 2) For complex mixture separation, the restored dual-source images achieve average PSNR and SSIM of 25.0049 dB and 0.7997, surpassing comparative methods by 4.1249 dB and 0.0926. Both subjective and objective evaluations confirm DCDSM's superiority in addressing rain/snow residue removal and detail preservation challenges.

Figures (8)

• Figure 1: Comprehensive architecture of the proposed DCDSM framework, illustrating the dual-branch diffusion process with interactive wavelet suppression modules
• Figure 2: Limitations of using conditional diffusion models alone for image separation, demonstrating the necessity of interactive suppression mechanisms
• Figure 3: Detailed architecture of the Wavelet Frequency-domain Feature Extraction Network (WFEN)
• Figure 4: Visual comparison of rain removal results on Rain-0.5 dataset, highlighting DCDSM's superior detail preservation and artifact suppression
• Figure 5: Visual comparison of snow removal performance across different methods