Skip to content

Khmer language

Extract Text from an Image with Free OCR (Khmer Too)

Updated: 6 min read

To extract text from an image, you need OCR (optical character recognition): software that recognises the letters in a picture and turns them into text you can copy. ToolOrca has three free OCR tools that do this in your browser: Khmer OCR for Khmer documents, Image to Text for 20 languages, and PDF OCR for scanned PDFs. They use Tesseract 5, a widely used open-source OCR engine, read Khmer together with English by default, and do not upload your files unless you choose the server option.

On clear images, around 98–99% of characters come out right in our tests. The rest of this guide shows how to get there, and how to check and clean up what OCR gives you.

What OCR is and when you need it

A photo of a page, a screenshot or a scanned PDF contains pixels, not letters. You can read it, but a computer cannot search it, copy from it or translate it until OCR has turned the shapes into characters.

You need OCR when:

  • You photographed a document, a sign, a whiteboard or a book page.
  • You received a scanned PDF and cannot select its text.
  • You have a screenshot of text that you want to edit.

You do not need OCR for most PDFs created on a computer. A PDF exported from Word, Google Docs or a website already has a text layer. PDF OCR notices this and takes that text directly, and PDF to Text gets it out too. Only pages that are pictures need to be read.

How to extract text from an image step by step

  1. Open the right page. For Khmer writing, open Khmer OCR. For other languages, open Image to Text. For a scanned PDF, open PDF OCR. All three use the same tool, with different starting settings.
  2. Add your files. Click Choose images or PDFs, drop files onto the box, paste a screenshot with Ctrl+V, or tap Take a photo on a phone. You can add several images or PDFs at once.
  3. Check the language and layout. On the Khmer page, leave the language on Khmer with Also read English words and numbers ticked. Keep the layout on Block of text for most pages; choose Automatic for columns and Single line for a sign or a label. For a PDF, open More options to choose PDF pages such as 1-3, 5.
  4. Click Extract text. The first time, the engine and the language data are downloaded once (Khmer is about 1 MB) and kept for next time. Each page is cleaned up automatically before reading: a tilted photo is straightened, small text is enlarged and uneven light is evened out.
  5. Check the result. Switch to Confidence view: words the engine is unsure about are marked yellow (worth a check) or red (probably wrong). Fix them in the text box.
  6. Copy or save the text. Use Copy text or Download TXT. With several files you can get One TXT file per file (ZIP), and for PDFs you can tick Also make a searchable PDF to keep the original pages with a hidden text layer.
  7. Check Khmer spelling. Click Check Khmer spelling to open the text in the Khmer Spell Checker and review each suggestion yourself.

Tip: By default everything runs On this device, so your files never leave your phone or computer. For a long PDF on a slow phone, switch to On our server: the prepared page images are read by the same engine and deleted right after the text is sent back.

Getting better OCR results for Khmer

Khmer is harder for OCR than English. Letters stack above and below each other, subscript consonants are small, and many fonts draw the same letter quite differently. The automatic clean-up helps a lot, but a good photo still matters most:

  • Use a sharp, large image. Hold the phone flat above the page, in good light, without your shadow on it, and fill the frame with the text. A screenshot is always better than a photo of a screen.
  • Crop away everything else. Tables, hands and the next page confuse the engine. Crop with Crop Image or take the photo closer.
  • Pick the matching layout. Multi-column pages read better with Automatic; a single line of text with Single line.
  • Try another clean-up mode when a result looks wrong. Black and white helps with faded print; Off is best for clean digital images such as screenshots.
  • Printed text rather than handwriting. The engine is trained on printed fonts, so handwritten Khmer gives a rough result at best, and decorative fonts such as Moul are read poorly.

On our mixed Khmer and English test images, OCR got about 2% of characters wrong; on clean scans and good phone photos, about 0.5–1%. That is very good, but on a long page it still means a few mistakes, so plan for proofreading.

Fixing OCR mistakes in Khmer text

The Khmer OCR tool already does a Khmer clean-up: vowel signs and subscripts are put in the standard Unicode order, doubled marks and stray spaces are removed. It does not change spelling on its own, because automatic correction tends to “fix” names and rare words that were right.

Typical mistakes are look-alike letters, a missing small vowel sign under a letter such as ុ or ូ, and a wrong subscript. To find them, press Check Khmer spelling and work through the Khmer Spell Checker:

  • Red words are not in its dictionary, which often means OCR misread a letter. Tap the word and pick a suggestion.
  • Orange words are in the wrong Unicode order. Fix Unicode ordering repairs the whole text in one go.
  • Blue words are real words that are often confused, such as និង and នឹង. Read them in context.

Our guide on checking Khmer spelling goes into more detail.

Tip: Old Khmer PDFs typed with Limon or ABC fonts give Latin gibberish instead of Khmer when their text is copied. Read those pages as images with OCR, or convert the copied text with Limon to Unicode.

Other options

If you are already in Google’s apps, Google Lens on a phone and Open with → Google Docs in Google Drive can also recognise text, including Khmer. They process your image on Google’s servers, and you get no confidence view, no Khmer clean-up and no searchable PDF, but they are a reasonable second opinion when one tool struggles with an unusual image.

Common mistakes to avoid

  • Running OCR on a PDF that already has text. Leave use the text already in the PDF ticked so digital pages are taken directly.
  • Photographing at an angle. Small tilts are straightened automatically, but a strongly skewed page still loses detail. Hold the phone straight above the page.
  • Trusting numbers without checking. OCR can read 1 as 7 or 0 as O. Check amounts, phone numbers and dates against the original.
  • Using OCR text for official documents without proofreading. A recognised contract or certificate must still be checked word by word by a person.
  • Choosing the wrong language. The engine only recognises the scripts you select, so Khmer text read as English comes out as nonsense.

Summary

ToolOrca now has free OCR: Khmer OCR for Khmer documents, Image to Text for 20 languages and PDF OCR for scanned PDFs, with page ranges, batch ZIP downloads and searchable PDFs. It runs in your browser by default, reads Khmer with English, cleans up each image automatically and gets around 98–99% of characters right on clear images. Check the yellow and red words in the confidence view, then finish with Check Khmer spelling in the Khmer Spell Checker.

Frequently asked questions

Does ToolOrca have an OCR tool?

Yes. Khmer OCR reads Khmer (with English) from photos, screenshots and scans, Image to Text reads 20 languages, and PDF OCR handles scanned PDFs and can make them searchable. All three use the same engine, Tesseract 5, and are free with no sign-up.

Are my images uploaded?

Not by default. With On this device the OCR engine runs in your browser, so your files stay on your phone or computer; only the engine and language data are downloaded once. If you choose On our server, the prepared page images are sent over an encrypted connection, read and deleted right after the text comes back.

How accurate is OCR for Khmer?

On clear printed text, such as a sharp screenshot, a clean scan or a good phone photo, around 98–99% of characters come out right in our tests. Blurry, small or tilted photos give more mistakes, and handwriting and decorative fonts such as Moul are not read well. Always proofread names, numbers and dates.

Do I need OCR to copy text from a PDF?

Not always. A PDF exported from Word or a website already contains real text. PDF OCR detects this and takes that text directly, which is faster and exact. Only pages that are pictures, such as scans, are read by OCR.

Can it read Khmer and English in the same image?

Yes. Khmer with Also read English words and numbers is the default on the Khmer OCR page, so names, prices and web addresses in a Khmer text are read too. Untick it only for text that is purely Khmer.

Why does copied Khmer text look right but not work in search?

The letters may be stored in the wrong Unicode order or contain invisible characters. The OCR tool already fixes vowel and subscript order; for text from other sources, use Fix Unicode ordering in the Khmer Spell Checker and the Invisible Character Remover.