.png){kind=icon}
.png){kind=icon}
Return
DiffuseDoc: Document geometric rectification via diffusion model
.png){kind=icon}
PRE
AI
DOI：10.1016/j.cviu.2025.104554.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose DiffuseDoc, a comprehensive framework for geometric rectification of document images. To the best of our knowledge, DiffuseDoc is the first geometric document rectification model that adopts the diffusion model. • We leverage the rectification result as the control condition and derive the diffusion process with a learnable condition. By employing the diffusion prior, we effectively control the distribution of the rectifying result. • We contribute a dataset. The dataset comprises real-world document images captured by sensors, along with their corresponding scanned versions. • Extensive experiments show that our approach achieved state-of-the-art performance on benchmark datasets and our collected datasets, surpassing other methods in various evaluation metrics.
Journal
IF:
3.5
Papers:
428
Citations:
7.3K