Add Deskew Mode Using Top Page Edge as Reference for Book Scanning

[hbenitezs]

Feature Request: Deskew Mode Based on Top Page Edge

Hello ScanTailor Advanced team,

I am part of a team processing scanned books for a university library. We currently use ScanTailor Advanced to process book scans, but we face an issue with the deskew functionality. The current deskew algorithm relies on internal page content (e.g., text lines or other features) to determine the rotation angle, which often leads to inconsistent results for pages with irregular content, such as illustrations, sparse text, or damaged pages.

We would like to request a new deskew mode that uses only the top edge of the page as the reference for determining the rotation angle, ignoring internal content. This would ensure that the page is aligned based on its physical border, which is critical for our book scanning workflow, as our scans typically have a clear top edge with a dark background.

Proposed Functionality:

• Detect the top edge of the page (e.g., by identifying the transition from the dark background to the lighter page content using edge detection or binarization).
• Fit a line to the detected top edge (e.g., using linear regression) to calculate the skew angle.
• Rotate the page to make the top edge horizontal, ignoring any internal content like text or lines.
• Add an option in the Deskew settings (e.g., a checkbox or dropdown) to select between the current content-based deskew and the new top-edge-based deskew.
• If no clear top edge is detected, fall back to the current algorithm or allow manual adjustment.

Use Case:
This feature would be particularly useful for book scanning projects where pages have consistent physical borders but varied content (e.g., books with illustrations, mixed text/image pages, or older books with irregular text alignment). In our case, we process scans for a university library contract, and aligning pages based on their top edge would greatly improve consistency and quality.

Kind regards

[Piolie]

Hi, this project had its last commit pushed on May 31, 2020, more than 5 years ago. I'd say it can be considered abandoned by now (sadly).

As alternatives, check the STA fork by Virgil Grigoras and some others at Image Processing for Electronic Publications.

All seem to be solo dev projects, but you could try opening an issue with either of them. I think it'd be a good functionality to have.

PS: Scan Tailor Universal is a different fork from the original Scan Tailor which has also seen some development, though I think the code base has diverged further than the others.