A novel vector-valued diffusion approach for optical image denoising guided by structure tensor

Impulse noise severely degrades optical images, affecting subsequent processing and analysis. This paper proposes a structure tensor-guided vector-valued diffusion model for effective noise removal while preserving structural and chromatic fidelity. Our method constructs a joint structure tensor to fuse multi-channel gradient information, enabling anisotropic diffusion along image edges. A two-stage noise detection algorithm generates a precise mask, restricting diffusion only to corrupted pixels. Experiments show the approach significantly outperforms state-of-the-art methods including adaptive median filter, total variation, and marginal diffusion models.