Amharic OCR: how to copy Amharic text from an image

How Amharic OCR works

OCR (optical character recognition) finds the letters in a picture and turns them into text. Amharic is harder than English: the Ethiopic script (ፊደል) has hundreds of characters, and many differ only by a small mark, such as ነ and ን, or ሰ and ስ.

Picsawa uses Tesseract, the open-source OCR engine, which has a trained model for Amharic (language code amh) and one for the Ethiopic script (Tesseract language list). On top of that, the tool:

• cleans up phone photos first, removing shadows and uneven light;
• reads mixed lines with both the Amharic and English models;
• re-reads words that came out as Latin letters inside Amharic lines;
• compares unknown words with a list of about 66,000 common Amharic words to fix typical slips such as ን/ነ, ስ/ሰ and አ/እ;
• highlights words it isn't sure about, so you know what to check.

In Picsawa's tests on printed Amharic in four common fonts, these fixes cut the number of wrong words by about a third. Clear fonts such as Noto Sans Ethiopic came out almost perfectly.

Step by step

1. Open Amharic Image to Text.
2. Add one or more photos or screenshots, or take a photo of the page. Rotate any page that is sideways, and put the pages in order.
3. Keep Amharic + English selected (use English only for English-only pages) and tap Read text. The first time, the engine and models (about 6 MB) are downloaded once; after that, it also works offline.
4. Read through the result. Highlighted words are the ones to check; edit them right in the text box.
5. Copy the text, or download it as a .txt or .docx file to keep working in Word or Google Docs.

Taking a photo that reads well

• Flat page, straight above. Lay the page on a table and hold the phone parallel to it; curved or angled lines read worse.
• Daylight, no shadow. Your hand and phone easily cast a shadow; stand so the light comes from the side.
• Fill the frame with the text. The bigger the letters in the photo, the better. Photograph long pages in two halves.
• Sharp focus. Tap the text on screen to focus, and hold still. A blurry ፊደል loses the small marks that tell characters apart.
• No flash on glossy paper: it causes a bright patch that wipes out letters.

Screenshots and PDFs

Screenshots are sharp, but small text on a phone screen has very few pixels per letter. Zoom in until the text is large before taking the screenshot, and take several if needed. For a PDF, zoom in and screenshot each page, or photograph the printed page. If the PDF already contains real text (you can select it), copy it directly instead.

Fixing common misreadings

Even with the fixes, check these spots:

• Look-alike characters: ነ/ን, ሰ/ስ, አ/እ, ለ/ል, and characters that differ by a small foot or ring.
• Punctuation: the Amharic full stop ። and comma ፣ can be read as dots or colons.
• Numbers: dates, phone numbers and amounts matter most; compare them with the original.
• Names and rare words aren't in the word list, so they're more likely to be wrong; they are often highlighted.

Correcting in the text box is quick. For long documents, download the .docx and use find and replace in Word or Google Docs to fix a word that was misread the same way many times.

What Amharic OCR can and can't do

• Good: printed letters, books, forms, notices, news articles, screenshots of chats and websites, and pages mixing Amharic and English.
• Harder: decorative and serif fonts, very small text, low-light photos, and columns or tables (the text may come out in a different order).
• Not suitable: handwriting, which OCR engines trained on printed text read poorly.

Everything runs on your device: documents are never uploaded, which matters for IDs, letters and contracts.