Yes, you can turn an image of Arabic, Chinese, Japanese, Korean, Russian or Hindi text into editable text online. Our OCR reads 29 languages, including Cyrillic, Greek, Arabic, Persian, Devanagari, Thai, Japanese, Korean and both Simplified and Traditional Chinese. The Document language menu on a tool like image to Word either stays on Automatic or takes the language you pick before you upload.
Automatic picks the recognition model from the script it finds on each page, so a Russian, Arabic or Japanese page is read with the right model without any setting. Choosing the language yourself, when you know it, can't do worse, and it's the way to get Traditional Chinese right. This guide shows how to set the language, what we got for each script, and where the output still needs fixing by hand.
How to Convert an Image to Text in Arabic, Chinese, Japanese or Russian
- Open an OCR tool: image to Word for an editable DOCX, PNG to TXT or JPG to TXT for plain text, scanned PDF to Word or searchable PDF for multi-page scans
- Choose the document language — the menu lists each language in its own script (العربية, 日本語, 한국어, Русский, हिन्दी, ไทย…), or leave it on Automatic, which recognises the script itself
- Upload the file. Recognition runs on our servers over an encrypted (HTTPS) connection
- Download the DOCX, TXT or PDF and check it against the original
Most of these pages take several files at once and queue them; a language you choose applies to all of them, while on Automatic each file gets its own detection. If one page mixes two scripts, pick the non-Latin one — see the section on mixed documents below.
Every OCR tool has the menu, including the "to text" pages that show the result on screen (PNG, JPG and PDF to text, plus the multi-image versions).
Which Languages the OCR Reads
The 29 options, grouped by script:
| Script | Languages in the menu |
|---|---|
| Latin | English, German, Spanish, French, Italian, Dutch, Polish, Portuguese, Turkish, Indonesian, Vietnamese, Czech, Swedish, Danish, Finnish, Norwegian, Hungarian, Romanian |
| Cyrillic | Russian, Ukrainian |
| Greek | Greek |
| Arabic script | Arabic, Persian |
| Devanagari | Hindi |
| Thai | Thai |
| CJK | Japanese, Korean, Chinese (Simplified), Chinese (Traditional) |
Not on the list: Hebrew, Bengali, Tamil, Georgian, Armenian and other scripts. The menu will offer only what it can read, so if your language isn't there, use a different tool rather than the closest neighbour — Hebrew set to Arabic won't work.
Our Test: One Sample per Script
We ran one image per script through PNG to TXT with the matching language selected. Each sample is clean, printed text — a heading and one line, set in Noto fonts and rendered at 200 DPI. That's an easy case: real scans with shadows, small print or low resolution will do worse.
| Script (language chosen) | Result |
|---|---|
| Latin — English, German | Exact, including ö, ß and Ü |
| Cyrillic — Russian | Exact, including ё |
| Cyrillic — Ukrainian | Exact, including the Ukrainian "і" |
| Greek | Exact, accents included |
| Devanagari — Hindi | Exact, conjuncts included |
| Thai | Exact |
| Japanese | Exact — kanji and kana in the same line |
| Korean | Exact |
| Chinese (Simplified) | Exact |
| Arabic | Exact, in right-to-left reading order |
| Persian | Exact, with the Persian letters "ی" and "ک" |
We also checked the searchable PDF tool for Russian, Japanese and Hindi: the invisible text layer matched the page, and searching the PDF for a word from the image found it. For Arabic, the text layer read back in the right order too.
Arabic and Persian OCR: Right-to-Left, in Reading Order
Arabic text comes out the way it reads, right to left, with every line in its place. We ran a four-line Arabic page through each output:
- Plain text (TXT): every word right, each printed line on its own line, in reading order
- Word (DOCX): the same four lines as paragraphs, ready to edit or paste into a translator
- Searchable PDF: the text layer held all four lines in the right order, so copying from the PDF gives the same text
Persian uses the same Arabic-script recognition, and our three-line Persian sample came back word for word. It keeps the Persian "ی" and "ک" rather than their Arabic look-alikes, so a search for a Persian word finds it.
Chinese: Pick Simplified or Traditional to Match the Print
Simplified and Traditional Chinese are separate menu entries, and it matters which one you choose. When we ran a Simplified sample (销售报告) with Traditional selected, the output came back in Traditional characters (銷售報告) — readable, but not what was printed. Copy that into a search box or a translator and you're working with different characters.
- Mainland China, Singapore: Chinese (Simplified)
- Taiwan, Hong Kong, Macau: Chinese (Traditional)
Automatic reads Chinese with the Simplified model, so for Traditional print choose Chinese (Traditional) yourself.
Our samples were horizontal. We haven't tested vertical columns (common in older books and newspapers), so check the output line order if your page reads top to bottom.
Japanese and Korean: Image to Text
Japanese mixes kanji, hiragana and katakana in one sentence, and all three came through correctly with Japanese selected. Korean Hangul was exact too. Two practical notes:
- Small print: furigana (the tiny kana above kanji) and dense text need a sharp source. If your photo is soft, rescan or re-shoot it closer — scan at 300 DPI, in even light, with the page flat
- Vertical Japanese (tategumi) wasn't part of our test, so treat it the same as vertical Chinese: check the order
Russian and Other Cyrillic Text
Russian was exact in our test, and so was Ukrainian, including the letter "і", which looks exactly like the Latin "i" but came back as the Cyrillic one, so searches for Ukrainian words match.
Serbian, Bulgarian and other Cyrillic languages aren't in the menu yet. Russian is the closest match for Cyrillic text, but letters that Russian doesn't use may come out wrong.
Mixed-Language Documents
On Automatic, the language is detected page by page, so a document whose pages are in different languages — an English cover followed by Russian pages, a French header page and Arabic pages — is read correctly without splitting it. A language you choose in the menu applies to every page of the job.
Within one page, the whole page is read with one model. In our test, the Russian, Japanese, Chinese and Arabic settings all read an English line perfectly, so a Japanese manual with English product names, a Russian letter with Latin addresses or an Arabic report with an English line is fine — pick the non-Latin language. On the Arabic page, the English line kept its own left-to-right word order between the Arabic lines.
Tips for Better Non-Latin OCR
- Resolution matters more than for Latin text. Strokes in Chinese, Devanagari conjuncts and Arabic dots are small. Scan at 300 DPI
- Mixed pages need no splitting. Automatic detects the language page by page; on a page that mixes scripts, pick its main language or keep the other-script part on its own page
- Check the font on your side. If the DOCX shows boxes instead of characters, the text is there but your font lacks the script — switch to Noto Sans or another font that covers it
- Printed text only. Handwriting in any script is a much harder problem
If the output looks wrong, check first that the document language matches the script or is set to Automatic, then that the page isn't a blurry photo; how online OCR works explains which step each problem comes from. And if you want to translate the text after extracting it, see OCR and translation.
OCR Languages From the Command Line or API
The fcl command-line tool takes a language code, so you can batch a folder of scans:
fcl convert scan.pdf --to txt --ocr-language ja
fcl convert *.png --to ocr-word --ocr-language ru --out-dir text/
Through the conversion API, OCR operations take a language option with the same codes as --ocr-language, such as ru, ja or ar.
Summary
For Arabic, Chinese, Japanese, Korean, Cyrillic, Hindi, Greek or Thai text, the one thing to remember is the Document language menu: set it, and our samples came back exact in every one of those scripts; leave it on Automatic, and it picks the script for each page itself — choose Traditional Chinese by hand. Start with image to Word or scanned PDF to Word.
Related Tools
- Image to Word (JPG) — editable DOCX with the language menu
- PNG to TXT and JPG to TXT — plain TXT files
- Scanned PDF to Word — multi-page scans
- Searchable PDF — keep the page image, add a text layer
- PNG to searchable PDF — for screenshots and single images