RecDiffusion: Rectangling for Image Stitching with Diffusion Models

TL;DR

RecDiffusion is a diffusion-based framework that rectanglizes stitched images by generating motion fields and refining details, improving geometric accuracy and visual appeal over previous methods.

Contribution

It introduces a novel diffusion-based approach combining motion and content diffusion models for rectangling stitched images, surpassing prior solutions in quality.

Findings

Outperforms previous methods in quantitative measures.

Ensures geometric accuracy and visual appeal.

Effective correction of irregular image boundaries.

Abstract

Image stitching from different captures often results in non-rectangular boundaries, which is often considered unappealing. To solve non-rectangular boundaries, current solutions involve cropping, which discards image content, inpainting, which can introduce unrelated content, or warping, which can distort non-linear features and introduce artifacts. To overcome these issues, we introduce a novel diffusion-based learning framework, \textbf{RecDiffusion}, for image stitching rectangling. This framework combines Motion Diffusion Models (MDM) to generate motion fields, effectively transitioning from the stitched image's irregular borders to a geometrically corrected intermediary. Followed by Content Diffusion Models (CDM) for image detail refinement. Notably, our sampling process utilizes a weighted map to identify regions needing correction during each iteration of CDM. Our RecDiffusion ensures geometric accuracy and overall visual appeal, surpassing all previous methods in both quantitative and qualitative measures when evaluated on public benchmarks. Code is released at https://github.com/lhaippp/RecDiffusion.