Arabic and Hebrew OCR: Right-to-Left Text Explained
Languages
Right-to-left scripts look intimidating to OCR, and historically they were. Arabic in particular changes letterforms depending on where a letter sits in a word, and its dots and diacritics are small features at scanning resolution. Hebrew has its own complexities, including a tradition of unvowelled text.
Modern vision models read printed Arabic and Hebrew well. The practical pitfalls are less about recognition and more about how the text behaves once you have it — here is what to expect.
Why Arabic is harder than it looks
Arabic letters connect, and most letters have up to four shapes: isolated, initial, medial, and final. The same letter is a different image depending on neighbors. Recognition therefore depends on word-level context, not just character shapes. Add the small dots that distinguish many letter pairs, and the cost of image blur multiplies.
Hebrew: letters and vowels
Hebrew letters are more uniform than Arabic's, and printed Hebrew recognizes well. The complication is vocalization (niqqud): small marks around letters that most modern Hebrew omits but prayer books and teaching materials include. Engines vary in whether they capture niqqud — if they matter, check a sample first.
The mixed-direction problem
Real RTL documents are rarely pure: they contain Latin brand names, English technical terms, and numbers. Numbers themselves are usually written left-to-right even in Arabic text. The result is bidirectional content, and the danger is at the edges — punctuation and short Latin runs sometimes land in the wrong order. Check quotes, parentheses, and dates in mixed paragraphs.
Reading order for RTL pages
Layout analysis must mirror its assumptions: in an RTL document, «first» means rightmost. Good engines detect the dominant direction from the page itself, not from a setting. On multi-column Arabic pages, column order should proceed right to left.
Practical tips
Scan at higher resolution — diacritics and dots are small features. Avoid lossy compression. For two-language documents, ask yourself which direction dominates and check the other side's fragments first: that is where ordering errors cluster.
Frequently asked
Does OCR output Arabic in correct visual order?
It outputs logical order, which is correct for text — your editor or browser then renders it right-to-left. If text looks reversed when pasted into an RTL-aware app, the problem is the app, not the OCR.
Are Arabic-Indic numerals preserved（٠١٢٣）?
Good engines preserve the original numeral forms. Check if your downstream system expects Western digits.
Can it read handwritten Arabic?
With effort. Modern models handle neat handwriting; cursive, fast handwriting remains difficult and needs verification.
Turn a page into text, free
Drop a photo, screenshot, or scanned PDF — get clean text in seconds.
Open the OCR tool