How to recognize English and Chinese text in a local image
Choose one local image, select English, Simplified Chinese or both, and recognize the previewed area. Tesseract runs in a cancellable browser Worker; core and language files load from this site only when recognition starts. Results remain editable so you can correct mistakes before copying or downloading.
How do you do it?
- Open one PNG, JPEG or WebP image up to 10 MiB, or load the sample for the selected language.
- Choose the language, rotation and crop. Check the exact recognition preview, then select Recognize text.
- Review and correct the result, then copy or download TXT. Clear the workspace or model cache when needed.
Recognize a clear English label
The sample uses large, horizontal printed text. This is a worked example, not a guarantee for photographs, handwriting or complex layouts. The selected image is processed locally.
Input:
A white image with black printed text:
LOCAL OCR 12345
Browser tools stay local
Output:
LOCAL OCR 12345
Browser tools stay localWhen is this workflow useful?
Use image OCR for printed text in screenshots, labels and clear scans. Use PDF Text Extractor for text already embedded in a PDF; this image tool does not rasterize PDFs. Use Image Resizer before recognition when a photo exceeds the pixel budget.
What should you check before relying on the result?
Single image, 10 MiB maximum. Pixel cap: 8 million on desktop and 4 million on devices with a coarse pointer. Crop is specified in percentages after rotation. Recognition stops after 90 seconds. No PDF input, handwriting guarantee, table reconstruction, camera capture or parallel batch OCR. Engine confidence is not verified accuracy. Only engine/model resources may be cached; images and text are not saved by this tool. English cold loading is about 4.3 MiB and combined English/Chinese about 5.9 MiB by gzip reference, not a speed or memory guarantee.
What can go wrong?
| Problem | What to try |
|---|---|
| No text or incorrect characters appear. | Small print, blur, skew, complex backgrounds or the wrong language make recognition less reliable. Choose the matching language, rotate the image to horizontal text and crop tightly around a clear text block. Review the output manually. |
| The image or engine cannot be loaded. | The image is unsupported, exceeds file/pixel limits, or static engine downloads fail. Use PNG/JPEG/WebP and resize first if needed. Check the connection and retry loading; a blocked model cache does not require uploading the image. |
Frequently asked questions
Do I need to upload data or sign in?
No. Processing runs in the current browser. This site serves the page and required static engine assets.
Why does the result disappear after I edit the input?
The tool clears stale output so a previous result cannot be mistaken for the new one. Run again after changing input or settings.
References
First edition: steps checked against the current tool interface, with examples and known limits.