Hi @Us3r369! I'm Dosu and I’m helping the docling team.

Docling doesn't have a built-in flag or API to directly tell you if a PDF is native (text-based) or scanned (image-based), but it does expose all the signals you need to implement this check in your pipeline. Docling's OCR pipeline uses bitmap (image) coverage per page: if a page is mostly image, OCR is triggered; if there's enough selectable text, OCR is skipped. This is controlled by options like bitmap_area_threshold (default 0.75) and force_full_page_ocr [source].

You can access the number of text cells and bitmap rectangles per page using Docling's PDF backends (e.g., PyPdfium2, DoclingParseV4), which lets you implement your own scan/native heuristic. For example, if a page has no text cells but large bitmap coverage, it's almost certainly scanned; if it has many text cells and little bitmap area, it's native [source].

If you want stricter or more explicit control, you can pre-process PDFs with a library like PyMuPDF (fitz) to check for text layers before running Docling, or analyze the DoclingDocument after conversion to infer the document type [source].

In summary: Docling doesn't provide a one-liner for this, but you can use its API to check for text cells and bitmap coverage per page and apply your own threshold or logic to decide if a PDF is scanned or native. This approach matches the heuristics you described and is how Docling internally decides when to apply OCR.

To reply, just mention @dosu.

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}