Multispectral image stitching based on diffusion model

Guangqiu Chen

doi:10.1016/j.image.2026.117752

Multispectral image stitching based on diffusion model

2026-10-28

0

PRE

AI

DOI：10.1016/j.image.2026.117752

Original

Save

Abstract

En 中文

• Unified diffusion-based stitching framework – improves SSIM by 8.5% and EN by 6.2% over cascade methods. • Mask-guided dual-modal independent alignment – reduces misalignment by 18% and artifacts by 23%. • Gradient-intensity joint loss with multi-channel fusion – boosts AG by 12.7% and PSNR by 1.2 dB. • Validated robustness and superiority – ranks first in 5/6 metrics, 15% higher success rate in cross-scene tests.

Keywords:

Diffusion models

Infrared-visible image stitching

Multi-modal fusion

Geometric alignment

Panoramic perception

Multispectral image stitching based on diffusion model

Abstract

Journal

Organization

Cited Papers