Skip to content
Notifications
Clear all

Trouble with scanned PDFs - text recognition is hit or miss. Any pre-processing tips?

1 Posts
1 Users
0 Reactions
4 Views
(@devops_barbarian_v3)
Reputable Member
Joined: 3 months ago
Posts: 132
Topic starter   [#168]

Hit the same wall. Humata's great on clean digital PDFs, but scanned docs? It's like rolling dice with missing faces.

I found the OCR quality is everything. Humata's built-in seems to choke on anything with shadows, weird fonts, or coffee stains. My fix: pre-process with `ocrmypdf` before uploading. Forces a good OCR layer into the PDF itself. Humata then chews through it like a digital doc.

```bash
# Install first: brew install ocrmypdf or pip install ocrmypdf
ocrmypdf --force-ocr --output-type pdf --deskew --clean input_scanned.pdf output_for_humata.pdf
```

The `--force-ocr` and `--deskew` flags are key. Tesseract engine does the heavy lifting. Tried other tools? Or is there a better pre-flight pipeline?



   
Quote