Converting Handwritten Notes to Searchable PDFs

Printed OCR is almost solved. Handwriting is still an open research problem, but today’s tools get you 85% of the way for most hands.

Why handwriting is hard

Printed characters live on a grid; handwriting slopes, loops and overlaps. Modern HTR (Handwritten Text Recognition) models are LSTM/Transformer hybrids trained on millions of lines; they handle neat cursive at ~90% character accuracy and messy scrawl at 65–75%.

Capture tips

• Scan at 400 dpi, 8-bit greyscale, with the lamp calibrated.
• Use a flat background sheet to kill shadows.
• If photographing, align the camera parallel to the page.

Preprocessing

1. Deskew.
2. Binarise with Sauvola or Niblack (adaptive thresholding handles paper tint).
3. Despeckle.
4. Line segmentation.

Tools we use

For high-stakes archives: Transkribus (Cloud, EU-hosted, trained models). For personal notes: Microsoft OneNote’s HTR. For developer workflows: TrOCR (Microsoft) or Kraken (open source).

Realistic expectations

Expect to proofread. For a 20-page notebook, budget 10 minutes of cleanup to reach 99%+ accuracy.

Workflow on PDFAPPLE

1. Scan to PDF.
2. Upload to the PDF to Word tool and enable the Handwriting engine (Pro plan).
3. Download the DOCX, fix remaining errors, re-export to PDF.

---

Tags: #handwriting OCR #scan notes #PDF notes #digital notebook

---