United Arab Emirates version. Set up for the UAE's bilingual paperwork: Arabic with English switched on, plus how the engine handled Arabic-Indic digits, joined words and right-to-left lines on test pages. US version
Arabic and English Image to Text (OCR)
Read Arabic and English from the same scan, such as a tenancy contract or trade licence page, in your browser, with tested tips on digits and joined words.
The interactive tool loads here when JavaScript runs.
About this toolWhat is the Image to Text (OCR) tool?
Arabic is the official language of the UAE, and the government portal notes that road signs and signs on commercial outlets are in Arabic and English. When a contract, licence or notice is laid out the same way, with an Arabic block and an English one, choose Arabic as the language and leave "Also recognise English" ticked. The page then loads both trained models (ara and eng), and the engine can draw on either for each line.
Leaving English off is an easy mistake to make. In our test on one six-line sample with three Arabic and three English lines, Arabic alone read the Arabic correctly but turned every English line into strings of symbols and stray Arabic letters, and confidence fell from 93% to 74%. With both languages loaded, all six lines came back in reading order.
Your scan stays on your device. The engine and the Arabic and English data files download from the jsDelivr CDN the first time, then the browser caches them, but the image itself is never uploaded. Even so, practise on documents without an Emirates ID number, passport details or bank information, and crop those areas out of anything you do read.
- OCR for PNG, JPG, WebP, GIF, BMP, pasted screenshots and scanned PDFs
- 17 languages including Chinese, Japanese, Korean, Arabic and Hindi
- Overall confidence plus a count of low-confidence words
- Automatic enlargement of small screenshots
- Copy to clipboard or download as a UTF-8 TXT file
- Images stay on your device; the engine and language data load once from a CDN
Updated
How to use the Image to Text (OCR) tool
- Drop images or a PDF onto the upload area, click Choose files, or paste a screenshot with Ctrl+V or Cmd+V anywhere on the page.
- Pick the language of the text. For a non-English document that also contains English words, keep Also recognise English ticked.
- Leave Layout on Automatic for documents. Choose Sparse text for UI screenshots and labels, or One uniform block for a cropped paragraph.
- For a PDF, optionally type the pages to read, such as 1-3. If it is password protected, a password box appears after the first attempt.
- Click Extract text. The first run downloads the engine and language data; progress is shown as each image or page is read.
- Check the confidence badge, correct anything that looks wrong directly in the text box, then click Copy or TXT.
Worked examples
A bilingual tenancy summary
Input
PNG, 700 x 330 px, three Arabic lines and three English lines; language Arabic, Also recognise English ticked, Automatic page layout, enlarge small images on
Output
All six lines read, confidence 93%. English lines exact. Arabic lines readable with three slips: "إيجار سكني" came out as one word, "تبدأ" lost its hamza shape, and "1 يناير" lost the space
The small image was enlarged 3x before reading. Joined words and dropped spaces are the typical Arabic errors on clean print, because the gap between words is narrow next to the joins inside a word. Search the output for long run-together words before you paste it anywhere official.
Rent written in Arabic-Indic digits
Input
Arabic lines with the rent as ٨٥٬٠٠٠ and a contract number ٢٠٤٧, then the same rent line with 85,000; Arabic plus English
Output
Confidence 82%. The Arabic-Indic amounts were misread as a mix of Latin letters and wrong digits; the line written with 85,000 came out exactly
The engine handled Western digits well and Arabic-Indic digits badly in this test. Retype every amount, date and reference number written in ٠١٢٣٤٥٦٧٨٩ from the original, and use the confidence figure only as a warning, not as a check.
The same page printed very small
Input
The rent sample rendered at about 9 px text, enlarge small images on
Output
Confidence 69%. "السنوي" became "الستوي" and the Arabic-Indic figures came out garbled
Arabic letters that differ only by dots, such as ن and ت, are the first to go at low resolution. Photograph or scan the page larger rather than relying on the enlarge option.
Background and insight
Tesseract was developed at Hewlett-Packard between 1985 and 1994, was released as open source in 2005, and was developed further with Google's sponsorship. Version 4 replaced the old character-shape classifier with an LSTM neural network that reads whole lines at a time, which made it far better on ordinary printed text. It first finds the page layout (blocks, paragraphs, lines), then recognises each line, then uses dictionary and language data to settle uncertain words. That is why picking the right language matters: German data knows about ß and long compound words, French data knows accented letters.
OCR quality is mostly decided before recognition starts. Tesseract's own documentation recommends images of at least 300 DPI and notes that accuracy falls when lower-case letters are under about 10 pixels high. A phone screenshot of 13 px interface text is right at that edge, so this tool can enlarge small images two or three times before reading them. Sharp focus, even lighting, straight lines and dark text on a plain light background matter more than any setting.
Every recognised word comes with a confidence from 0 to 100. The tool reports the average, weighted by how much text each image held, and counts words under 60. Treat it as a hint rather than a measurement: Tesseract can be confident and wrong, especially with look-alike characters such as 0 and O, 1 and l, or rn and m.
Common use cases
Copying text from screenshots
Grab an error message, a code snippet in a video frame or a paragraph from an image someone sent you.
Digitising paper
Turn photos of printed letters, book pages, menus or handouts into searchable, editable text.
Scanned PDFs
Get the text out of a PDF that came from a scanner, which PDF to Word converters cannot read.
Receipts and invoices
Pull totals, dates and reference numbers from receipts into a spreadsheet or expense report.
Translation
Extract text from a photo of a sign or document in another language so you can paste it into a translator.
Tips and limitations
- Keep "Also recognise English" ticked for UAE documents unless the page is Arabic only; it costs a slightly longer first download and protects every English line, name and email address.
- English lines in the result are wrapped in invisible left-to-right and right-to-left marks (U+200E and U+200F). They keep mixed text in order when pasted into a word processor, but can break an exact search or a spreadsheet match, so strip them if a lookup fails.
- Check Arabic words that look too long: two words run together is the most frequent slip.
- Proofread hamza and dotted letters (أ, إ, ة, ن, ت, ي) in names and addresses, where one wrong dot changes the word.
- Handwritten Arabic, signatures and stamps are not read in any useful way; type those parts yourself.
Sources and further reading
- Fact sheet: official language, The UAE Government portal. Arabic as the official language; signs in Arabic and English.
- tessdata: trained models (ara, eng), Tesseract OCR project
- Improving the quality of the output, Tesseract documentation
- tesseract.js, Project Naptha
Frequently asked questions
Can this OCR read Arabic text from an image?
Yes. Choose Arabic in the language list. It reads printed Arabic with the Tesseract engine running in your browser; handwriting is not supported.
How do I extract text from a document in both Arabic and English?
Choose Arabic and keep "Also recognise English" ticked. In our tests on sample pages, leaving English off garbled every English line on a bilingual page and cut confidence from 93% to 74%.
Does it read Arabic-Indic numerals correctly?
Not reliably. In our tests on sample pages, amounts and reference numbers written in Arabic-Indic digits were misread, while the same amount in Western digits came out exactly. Retype those numbers from the original.
Is it safe to scan my Emirates ID or tenancy contract here?
The image is processed on your device and not uploaded; only the OCR engine and language files are downloaded. Still, crop out ID numbers and personal details you do not need, and do not share the extracted text more widely than the original.
Why does the Arabic text look jumbled when I paste it?
Arabic runs right to left, and the output includes direction marks around English lines to keep mixed text in order. If an app shows it oddly, paste into a document set to right-to-left, or remove the U+200E and U+200F marks for plain data work.
Can it read a scanned PDF trade licence?
Yes. Each PDF page is rendered at 300 DPI and then read, and multi-page results are joined with a page marker. Check the licence number and dates by hand.
Is there an in-depth guide to the Image to Text (OCR) tool?
Yes. “How OCR Works and How to Get Accurate Text From Images” covers the background, how it works, common pitfalls and worked examples. Read the full guide.
Related Tools You May Like
Free, in-browser tools that pair well with this one.