I think I have identified the cause of this issue and found a workaround that works well enough.
I was taking image files (jpg and png) and running them through DT’s OCR. It seems like the deskew function was tweaking the size of the images as it imported them, sometimes wildly. I guess this means that deskew cannot be disabled for images… perhaps this is a bug? On the other hand, I find that the deskew function does turn off when running OCR on PDFs.
What I do now is three steps:
- First, I batch process the files in a photo editing program. I use Affinity, it’s good enough for this kind of work. I adjust, compress, and set a uniform fixed width through its macro interface.
- Second, I convert to PDF in DT.
- Finally, I run OCR with deskew turned off (via settings).
This produces consistent document sizes and saves time in the end since I don’t have to redo work or get into nitty gritty editing. With a cropping step, this approach also yields good results for documents I captured with a camera or my phone—which is the only option in some archives.
The one downside is that without deskew, the OCR is less able to read words or form lines, particularly with handwriting.
I hope this is helpful to someone in the future.