How to Extract Text From an Image: A Practical OCR Guide (12 Languages)

The short answer

To extract text from an image, open the free Image to Text tool, choose the language of the text, and drop in a JPG, PNG or WebP. Recognition runs in your browser with Tesseract, an open-source OCR engine, and you can copy the result or download it as a .txt file. Clear, straight, well-lit images of printed text give the best results.

What OCR actually does

Optical character recognition looks at the pixels in an image, finds shapes that look like letters, and matches them against patterns it has learned for a language. That's why picking the right language matters: Hindi text recognised with the English model comes out as random Latin characters, because the engine is trying to match Devanagari shapes against English letter shapes.

The tool supports 12 languages: English, Hindi, Bengali, Tamil, Spanish, French, German, Portuguese, Russian, Arabic, Chinese (Simplified) and Japanese. The first time you use a language, its data file downloads once. After that, recognition runs entirely on your device, and your image is never uploaded.

Step by step

Choose the language of the text in the image.

Drop the image onto the tool or click to browse. JPG, PNG and WebP work.

Wait for recognition. A progress bar shows how far along it is. A typical phone photo of a page takes a few seconds on a recent laptop and longer on an older phone.

Check the confidence score. The tool shows Tesseract's average confidence for the page. Above roughly 85% usually means clean text; below 60% means you should expect corrections.

Copy or download the text.

Seven tips that improve accuracy

Most bad OCR results come from the image, not the engine. These fixes make the biggest difference, roughly in order of impact.

1. Shoot straight on. Text photographed at an angle has letters of different sizes across the page. Hold the phone parallel to the paper.

2. Use even light. Shadows across the page (often your own hand or phone) split letters into pieces. Daylight from a window, without direct sun, works well.

3. Fill the frame with the text. OCR works on pixels. A paragraph that takes up a fifth of a 12-megapixel photo has fewer pixels per letter than the same paragraph filling the frame.

4. Crop out everything else. Table edges, keyboards and photos on the page confuse layout detection. Use the Crop Image tool first.

5. Keep the resolution. Don't send the image through a messaging app first; WhatsApp and similar apps compress photos and soften letter edges. If you must compress for storage, do it after OCR.

6. One language per image. Mixed English and Hindi on one page is common in Indian documents. Run the image once with each language and combine the parts you need.

7. Printed beats handwritten. Tesseract is trained mostly on printed text. Neat block capitals often work; joined handwriting usually doesn't.

What it works well on

Source	Typical result	Notes
Screenshots of apps and websites	Excellent	Sharp, high-contrast text
Printed documents and books	Very good	Shoot straight and well lit
Receipts	Good	Thermal paper fades; photograph soon
Business cards	Good	Crop tightly; fancy fonts may need fixes
Signboards and photos of scenes	Mixed	Angles and backgrounds reduce accuracy
Handwritten notes	Poor to fair	Only very neat print works reliably

Scanned PDFs

If your text is inside a scanned PDF rather than an image, use PDF OCR instead, or convert the pages with PDF to JPG and run each page through the image tool. A PDF that already contains selectable text doesn't need OCR at all; copy the text directly.

Common errors and how to fix them

Letters swapped for similar shapes. "rn" read as "m", "0" as "O", "1" as "l". These are classic OCR confusions. Proofread numbers carefully, especially in invoices and IDs.

Words run together or split. Usually caused by uneven spacing or a slightly rotated image. Straighten the photo with an image rotate tool and try again.

Columns merged into one line. OCR reads across the page. For two-column layouts, crop each column and run them separately.

Confidence is low on a clean image. Check that you selected the right language.

Privacy: why browser-based OCR matters

Many free OCR websites upload your image to a server. That's a real concern for ID cards, bank statements, medical reports and contracts. With browser-based OCR, the image stays on your device; the only download is the language data file, and nothing is sent back. If you're extracting text from an Aadhaar card or similar ID, consider masking sensitive numbers first with the Aadhaar Mask tool.

OCR on common documents

Document	What usually works	Watch out for
Invoices and bills	Totals, dates, GSTIN	Table columns merging; crop the item table separately
Printed forms	Typed fields	Handwritten entries, checkboxes
Marksheets and certificates	Names, subjects, marks	Watermarks and seals over text
Book pages	Paragraph text	Curved pages near the spine; press the book flat
Slides and posters	Large headings	Stylised fonts and text over photos
Menus and signboards	Short phrases	Decorative fonts, angles and glare

Checking the output

OCR text always needs a proofread before you rely on it. A quick routine:

Numbers first. Amounts, dates, phone numbers, account and ID numbers are where a single wrong character matters most. Compare them with the image one by one.

Proper nouns. Names of people, places and companies are often outside the language model's vocabulary.

Line breaks. OCR keeps the line breaks from the image. To turn them into normal paragraphs, paste into a text editor and join lines, or use the Find and Replace tool.

Punctuation. Commas and full stops are small; on low-resolution images they can disappear or turn into other marks.

When OCR isn't the right tool

• The text is already selectable. If you can highlight text in a PDF or web page, copy it directly; it will be exact.
• You need the layout, not just the words. OCR returns plain text. To keep a document's look, convert the PDF instead with PDF to Word.
• You need tables as tables. For bank statements in PDF form, the Bank Statement to Excel tool keeps rows and columns.

Using the text afterwards

• Paste into Word or Google Docs for editing.
• Check spelling and grammar with the Grammar Checker.
• Count words for a submission limit with the Word Counter.
• Compare two versions of extracted text with the Text Diff Checker.