A scanned Vietnamese PDF is usually a stack of page images. You can see the text, but you cannot select it, search it or paste it into an email. OCR (optical character recognition) reads the shapes of the letters and turns each page into editable text. This guide shows the browser-only workflow, the language settings that matter for Vietnamese, and the limits to check before you start.
Before you OCR: source quality
OCR accuracy depends more on the scan than on the language model. A clean 300 DPI grayscale scan of a printed page usually gives the best result. Phone photos can work if the page is flat, evenly lit and in focus. A 150 DPI fax-style scan with speckles will produce more errors, especially with Vietnamese diacritics.
Check the first page before processing the whole file. If the text is skewed, crop and rotate it first with the Crop PDF and Rotate PDF tools. A straight page is easier for Tesseract to segment into lines.
If the PDF already has a text layer, do not OCR it. Use PDF to text instead; it extracts the existing characters without guessing. OCR is for image-only PDFs and photos.
OCR a Vietnamese PDF in five steps
- Open the OCR PDF & images tool in a desktop browser. Chrome, Edge or Firefox with enough free memory is the safest choice for a 20-page file.
- Drop the PDF, or add JPG, PNG or WebP photos of the pages. Wait for the page thumbnails to appear before changing settings.
- Set the OCR language to Vietnamese (vie). Add English (eng) only when the page really contains English words, such as invoices with product names.
- Start OCR and keep the browser tab open. The first run downloads the language model, so it can take longer than later runs. Progress is shown per page.
- Review the result, then download the TXT file. The tool exports plain text in page order; it does not rebuild the PDF layout.
For a mixed-language contract, OCR once with both languages rather than running the file twice. Running twice doubles the work and gives you two files to reconcile.
Vietnamese, English or both?
Tesseract uses separate language models. The Vietnamese model (vie) understands characters such as ă, â, đ, ê, ô, ơ, ư and the six tones. The English model (eng) is faster but will drop or misread those characters. Selecting both tells Tesseract to consider words from either model; it is useful for bilingual documents but slower and slightly more prone to false matches.
Use vie only for government forms, Vietnamese invoices, letters and books. Use vie + eng for software screenshots, import/export documents, or a page where product codes and English headings are mixed with Vietnamese sentences. Do not select a long list of languages “just in case”; each additional model adds processing time and ambiguity.
Improve accuracy without rescanning
If the output has broken words, check whether the scan is rotated by a degree or two. Straightening the page often fixes more errors than changing any OCR setting. For phone photos, crop out the desk or hand and keep only the page; extra background confuses line detection.
For very small print, upscaling the image before OCR can help, but do not sharpen aggressively. Sharpening creates halos around Vietnamese diacritics and can turn a clear mark into a wrong tone. A clean grayscale image at 300 DPI is a better starting point than a heavily filtered one.
Always review names, numbers, dates and legal terms. OCR can confuse similar-looking Vietnamese words, especially when the scan is faint. Treat the TXT as a draft, not as a certified copy.
Page, file-size and browser limits
Each OCR run accepts up to 20 pages, 100 MB per file and 300 MB total. A 21-page PDF will not start; split it first with Split PDF, then OCR the parts. Splitting also makes it easier to find the page where recognition failed.
OCR is memory-heavy. On a phone or a low-RAM laptop, process 5–10 pages at a time instead of the maximum. Keep the screen on and avoid switching apps while the worker is running. If the browser reloads, the job is lost and you need to start again.
The output is a TXT file. If you need a searchable PDF with the original page images, this tool is not the right output format; it gives you the text layer only. You can keep the original PDF for reading and use the TXT for search, translation or editing.
How to verify the file is not uploaded
Open the browser DevTools Network tab, filter by “Fetch/XHR”, then start OCR. You will see requests for the page, scripts and the Tesseract language model, but not a multipart upload of your PDF. The language model is a static file served from the same site; the document stays in browser memory.
You can also disconnect from the network after the tool page and language model have loaded. OCR will keep working because it does not call a recognition API. This is useful for confidential documents on a laptop that must stay offline.
Do it now
Open the OCR tool, choose Vietnamese, and process the first page before committing the whole file.
Related links
Frequently asked questions
Can PixToPDF OCR Vietnamese text without uploading the PDF?
Yes. The OCR tool runs Tesseract in your browser. The PDF or image is read locally and the Vietnamese language model is loaded as a static file; the document itself is not sent to a server.
Which language should I select for a Vietnamese scan?
Select Vietnamese (vie). If the page mixes Vietnamese and English, select both vie and eng. More languages increase processing time and can introduce more false matches.
How many pages can I OCR at once?
Up to 20 pages per PDF, 100 MB per file and 300 MB total per run. A longer PDF can be split first, then each part OCRed separately.
Does OCR keep the original layout?
No. OCR returns plain text, not a reconstructed page. Tables, columns and text boxes may be read in the wrong order, so review the TXT output before using it as a legal or financial record.