How to extract text from an image
Load one JPG, PNG, WebP, BMP, or GIF, then choose Language(s) and correct Rotation before you click Extract. You can draw a Crop region (optional) around the part that contains text. The result appears in Paragraphs, Lines, or Plain text, with confidence information available for review.
Image to text with language, rotation, and crop
Load one JPG, PNG, WebP, BMP, or GIF image. Use Drop an image here or click browse file. The file summary shows the image name, type, and pixel dimensions.
- Drop an image here or click browse file.
- Set Language(s), using Ctrl-click or Cmd-click when more than one language is present.
- Choose Rotation to correct sideways or upside-down text.
- Drag across the preview for a Crop region (optional), or leave it clear to recognize the whole image.
- Click Extract, then review the result.
Rotation offers None (0°), Rotate 90° clockwise, Rotate 180°, and Rotate 270° clockwise (90° counter-clockwise). Changing Rotation clears the current crop. Clear selection returns recognition to the whole rotated image.
Convert image to text and check confidence
OCR reads printed or typed running text and reports an overall confidence value. Choose Confidence threshold at 90%, 80%, 70%, or 60%. Words below that value are highlighted as uncertain in Paragraphs and Lines after recognition. Changing the threshold updates the highlighting.
receipt.png
Invoice #1042
Total: €42.50The same result can be viewed as a paragraph, as separate lines, or as plain text. Plain text has no confidence highlighting.
What the OCR exports include
Copy text sends the plain recognized text to the clipboard when clipboard access works. Download text (.txt) saves the same text. Download JSON records the source name, selected languages, overall confidence, threshold, paragraph and line data, per-word confidence, uncertainty flags, and word positions.
Download searchable PDF creates a one-page PDF from the rotated and optionally cropped image, with an invisible text layer over it. Characters that the PDF font cannot place can be left out of that layer while the page image remains.
What image-to-text OCR does not preserve
Paragraphs and Lines describe running text. They do not rebuild table rows, columns, merged cells, or form structure. Handwriting is not reliably supported. Recognition quality varies by language, so inspect words marked uncertain before copying the result into another document.
Examples with receipts, pages, and translations
- For an expense record, load
receipt.png, crop away the desk around the receipt, and download the text. - For a sideways page, set Rotation to Rotate 90° clockwise and review Lines before copying a quotation.
- For a translation, select two entries in Language(s), run Extract, and check low-confidence words in Paragraphs before saving JSON.
When to use Extract Text From Image
Use Image Extract Tools when the source is an image and you need OCR controls for language, orientation, or a selected region. Extract Text From Image returns running text, so table rows and columns will not stay structured in the result.
What happens to my image during OCR?
The image is read into browser memory, drawn to a local canvas, and passed to the in-browser OCR library. The script does not send the image to a server. An unreadable or corrupt file can stop loading. OCR or searchable-PDF export can also show an error if its library or operation fails.
The code does not define a file-size limit, maximum image dimension, processing-time guarantee, or browser compatibility matrix.
Frequently asked questions
How do I extract text from an image?
Load a supported image, set Language(s), adjust Rotation or Crop region (optional), and click Extract. Review Paragraphs, Lines, or Plain text before copying or downloading the result.
Can I convert an image to text in more than one language?
Yes. Select more than one entry in Language(s) with Ctrl-click or Cmd-click, then run Extract. Recognition quality varies by language, so check the confidence information.
What does the confidence threshold change?
It sets the percentage below which recognized words are highlighted as uncertain in Paragraphs and Lines. The available choices are 90%, 80%, 70%, and 60%.
Does image-to-text OCR preserve tables or handwriting?
The result represents running text in paragraphs and lines. Table structure is not reconstructed, and handwriting recognition is not reliable.
Is my image uploaded during OCR?
The image stays in browser memory while the page processes it locally. The script does not send the image to a server.