I add text layers to my scanned PDFs with a Google API–based OCR workflow of my own, outside DEVONthink, and have therefore turned off Settings › Files › Import › “Make text in PDF documents searchable.” The index behaves as expected: database search finds only the text that the files themselves contain.
The PDF viewer, however, still offers Apple’s Live Text on pages without a text layer (scans, scanned figures on typeset pages): the recognized text can be selected, and Edit › Find matches it within the document. The recognition is of low quality, and it blurs a distinction I rely on, namely between what a file contains and what the viewer infers from its images.
Is there a setting, or a hidden preference, that disables this recognition in the viewer? If it is PDFKit’s own behavior and lies beyond DEVONthink’s control, I would welcome a word to that effect; otherwise, please consider this a feature request.
That DT uses PDFKit has been established often here in the past. And since the developers occasionally still mention they have to work around its quirks, they most probably still rely on PDFKit.
Perhaps this 4 year old post is still useful:
Not to put too fine a point on it, but if you have an image and run OCR on it, you don’t do anything fundamentally different from Apple’s Vision framework. OCR adds a text layer, which Vision does not. And your OCR may be better than Apple’s. But technically, your file does “contain” the text only because you’ve added it before.
Thanks for the link, @chrillek. I’d rather not switch it off systemwide because, like the author of that post, I like the convenience of selectable text in screenshots. It is in DEVONthink specifically that I find Vision intrusive, because the PDFKit viewer tends to give it priority over embedded OCR when selecting and copying text. (This is arguably a PDFKit bug.) I have not ascertained how consistently it does so. Indeed, PDFKit is quirky.
(And yes, Google’s OCR is substantially superior to Apple’s and ABBYY’s—I labored over this evaluation and did it systematically, hence my confidence in it.)