How to Extract Nepali Text from Images (OCR Tips)

Retyping a printed Nepali notice, an old document or a screenshot by hand is slow. OCR (optical character recognition) reads the text from the image for you. It works well, if you give it a good image. This guide explains how Nepali OCR works and how to get clean, accurate results.

How OCR reads Nepali

Our Image to Text tool uses Tesseract, a widely used open-source OCR engine. Tesseract cleans up the image, finds lines of text, and reads each line with a neural network trained on printed text in that language. For Nepali it uses a model trained on Devanagari script, and the output is standard Unicode text that you can paste anywhere.

Devanagari is harder for OCR than English:

• The headline (शिरोरेखा) joins the letters of a word, so the engine can't simply split letters at gaps.
• Matras sit above, below and beside letters: ि ी ु ू े ै and the chandrabindu ँ are small marks that are easy to miss in a blurry image.
• Conjuncts like क्ष, त्र, द्ध and श्र combine two or more letters into one shape.
• Some letters look alike at low resolution: ब and व, म and भ, घ and ध, ङ and ड.

The clearer the image, the less the engine has to guess.

7 tips for accurate results

1. Use enough resolution

Text should be clearly readable when you zoom in. For phone photos of documents, fill the frame with the page. For screenshots, capture at normal or larger zoom. Very small text in a large photo is the most common cause of poor results.

2. Keep the page flat and straight

Photograph from directly above, not at an angle. Curved pages, like a thick book's spine, bend the lines and confuse line detection.

3. Use even lighting

Daylight near a window works well. Avoid shadows from your hand or phone, and avoid flash glare on glossy paper.

4. Crop to the text

Remove backgrounds, photos and decorative borders before running OCR. Our Crop Image tool makes this quick.

5. Choose the right language

Select नेपाली for Nepali-only text and English for English-only text. Use नेपाली + English only when the image really contains both. A single language is faster and makes fewer mistakes.

6. One column at a time

Newspapers and forms with several columns or tables may come out in the wrong order. Crop each column separately for clean paragraphs.

7. Printed text only

OCR engines like Tesseract are trained on printed fonts. Handwriting, highly decorative fonts and text over busy pictures will give poor results.

Proofreading checklist

Even with a good image, always check the result before using it, especially for official documents. Look for:

• Short and long vowels: ि vs ी, and ु vs ू.
• Chandrabindu vs anusvara: ँ vs ं.
• Look-alike letters: ब/व, म/भ, घ/ध.
• Numbers: Nepali digits (१२३) and English digits (123), especially in dates, amounts and phone numbers.
• Punctuation: the danda । sometimes comes out as | or 1.
• Line breaks: OCR keeps the line breaks from the image, so join lines into paragraphs if needed.

The text box in our tool is editable, so you can fix mistakes before copying.

Getting the text in Preeti

OCR always produces Unicode text. If you need Preeti, for example to paste into a document that uses Preeti, click Copy as Preeti in the Image to Text tool. It converts the text with our Unicode to Preeti converter. Paste it into your document and set the font to Preeti. Read Unicode vs Preeti to understand why this conversion is needed.

Reading text from PDFs

Scanned PDFs are just pictures of pages, so you can't copy text from them. The Image to Text tool reads every page of a PDF in order and labels the text by page number. If you can already select text in a PDF, it contains real text, and copying it directly will be more accurate than OCR. If you get English letters instead of Nepali, the PDF was made with Preeti: paste it into the Preeti to Unicode converter instead.

Privacy

Text recognition runs inside your browser. The first time, your browser downloads the OCR engine and language data. After that, your images are processed on your device and never uploaded.