OCR a Chinese PDF

Chinese OCR is harder than Latin OCR: thousands of characters with similar shapes, no spaces to help the model segment, and a lot of scanned documents that were photocopied twice before being scanned.

OCR Scanned PDF — free and local

Recognises text in scanned PDFs — English, Chinese or both — with a confidence score.

Use it now

✓ Nothing is uploaded ✓ No watermark ✓ No signup

简体 or 繁體 — and why the choice matters

The two models are trained on different character sets. Feeding a simplified document to the traditional model produces a recognisable but wrong output, full of characters that look similar and mean something else. Pick the script the document is actually printed in.

For a document that mixes Chinese body text with English brand names, model numbers or URLs, use “Chinese + English”. Two models run in sequence, which roughly doubles the time per page but noticeably improves the Latin fragments.

Why photocopied documents fail

OCR needs clean edges. A second-generation photocopy has broken strokes and grey noise, and Chinese characters depend on stroke detail far more than Latin letters do — a missing horizontal stroke turns one character into another.

If you control the source, scan at 300 DPI in greyscale, straight, with no colour cast. The tool applies contrast stretching automatically, but it cannot recover detail that was never captured.

Checking the result efficiently

Read the low-confidence pages, not all of them. A page at 95% usually has one or two wrong characters; a page at 60% usually has a systematic problem — a rotated scan, the wrong model, or a table that broke the line detection.

Names, numbers and dates are where OCR errors hurt most, and they are the easiest to verify against the page image. Start there.

Questions

Which model should I choose?

简体中文 for mainland documents, 繁體中文 for Taiwan/Hong Kong documents, and Chinese + English when the page mixes scripts.

Does it upload my document?

No. Recognition happens in your browser. The language model (about 2.4 MB for Chinese) downloads once and is cached.

Can it read handwriting?

Poorly. Tesseract is trained on printed text; handwritten Chinese is close to unusable. It will produce something, and you should not trust it.

Turn a scanned PDF into text Compress PDF — Shrinks a PDF with four presets — lossless cleanup, light, balanced and extreme Compress to Size — Takes your upload limit in KB or MB and finds the best quality that still fits under it Merge PDF — Joins several PDFs into one document, in whatever order you choose Split PDF — Splits one PDF into page ranges or a single file per page, delivered as a ZIP

SlimPDF runs every tool in your browser. No file is ever uploaded, stored or shared.