ZS-KAN: Zero-Shot Image Denoising With Lightweight Kolmogorov--Arnold Networks

Current learning-based image denoising methods have achieved impressive performance. However, their reliance on deep neural architectures and large datasets limits their applicability in data-limited or edge-computing scenarios. Therefore, we aim to develop a lightweight, effective, and efficient neural network for self-supervised image denoising.

Motivated by the functional approximation capability of Kolmogorov-Arnold networks (KANs), here we present ZS-KAN, a zero-shot denoising approach built upon a hybrid architecture that combines the computational efficiency of convolutional neural networks with the representational flexibility of KANs.

Experiments on synthetic and real-world noisy data show that ZS-KAN achieves performance comparable to or even surpassing state-of-the-art zero-shot denoising methods while using only 1% -25% of their parameter counts, significantly reducing model complexity. Visual comparisons indicate that ZS-KAN typically preserves more fine details than competing methods.

This study demonstrates a strong potential of KANs as a powerful component for zero-shot image denoising.

The combination of lightweight architecture, self-learning, and computational efficiency leads to the merits of ZS-KAN for real-world deployment, particularly in medical imaging workflows.