Image OCR は何をしますか？
Image OCR はブラウザ上でデータを調査するツールで、結果と制限を明確に示します。 検証済みの機能： English and Simplified Chinese models, Rotation and percentage crop, Editable text and TXT export, Local model cache and hard cancellation.
When should you use this tool?
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.
使い方
- 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.
実際の入出力例
Recognize a clear English label
入力
A white image with black printed text:
LOCAL OCR 12345
Browser tools stay local出力
LOCAL OCR 12345
Browser tools stay localThe 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.
一般的なエラーと対処
| 症状 | 原因 | 対処 |
|---|---|---|
| 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. |
Supported behavior and limitations
このツール固有の制限： 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.
標準と実装の参考資料
- Tesseract.js official repository (opens in a new tab)Pinned local Worker and LSTM engine, Apache-2.0. The site integration owns the Worker so initialization can be cancelled.
- Official Tesseract language data (opens in a new tab)Pinned English and Simplified Chinese 4.0.0_best_int models are hosted by this site. Model selection and image quality affect results.
Image OCR — FAQ
Is my input uploaded?
No. Processing happens in a browser Worker. Engine files load from this website; tool input, output, and filenames are not sent to a processing API.
Do I need an account or API key?
No account, API key, or backend processing service is needed.
How do I save a result?
Review the result, then use the Copy or Download control available for that output. Modifying input clears stale results.
What are the limits?
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.
更新内容
実質的な更新： Image OCR: Added local processing, explicit size limits, cancellation, worked examples, error explanations, and downloadable results.