Text Extractor (OCR)

The Text Extractor uses OCR (optical character recognition) to turn the text in a picture into text you can copy and edit. Choose the recognition language, upload a screenshot, photo or scan, wait a moment, and the result appears in a text box below. The engine is the open-source Tesseract, running in your browser through tesseract.js. OCR output is never guaranteed to be perfect, especially with blurry images, unusual fonts or complex layouts, so always proofread the result.

How to use it

1. First pick the main language of the image in the "Recognition language" dropdown. The options are Chinese + English (default), English only, Traditional Chinese, Japanese and Korean. The dropdown is locked while recognition is running.
2. Click the dashed area "Click to upload an image (screenshots, photos, or scans all work)" and choose a picture. Recognition starts immediately; there is no extra button, and only one image is handled at a time.
3. A preview of your image appears, along with a "Recognizing… N%" message. The first time you use a language, the files it needs have to be downloaded first, so the progress may sit still for a while.
4. When it finishes, a box titled "Recognized text (editable)" appears. Correct any mistakes directly in it, then select and copy the text elsewhere.
5. To process another picture, click the upload area again. To change language, change the dropdown first and then upload the picture, because the language is applied at upload time.

A practical scenario

Say you receive a photo of a whiteboard from a meeting and want the key points as text. Set the language to Chinese + English if the board mixes both, upload the photo, and wait for the progress to complete. Fix the few misread characters directly in the result box, then copy the text into your document. If the board had glare or untidy handwriting, expect many errors or even an empty result. In that case, retaking a straighter, brighter, sharper photo usually saves more time than correcting the output word by word. Printed text and clear screenshots generally work far better than handwriting.

Supported range and known limits

• There are only five language choices: Chinese + English, English only, Traditional Chinese, Japanese and Korean. Other languages such as French, German or Russian are not offered, and you cannot combine languages beyond these options.
• One image at a time. There is no batch mode and no PDF support. The output is plain text with no table structure, colours or fonts, and there is no copy or download button, so you select the text in the box and copy it yourself.
• Accuracy depends on the picture. Sharp, high-contrast, horizontal text works best. Blur, tilt, glare, low resolution, decorative fonts, handwriting and busy backgrounds cause errors. The tool offers no pre-processing such as straightening or contrast boosting, so prepare the image beforehand if needed.
• If no text is found in the image, the page shows a "No text was found" notice. Try another image or check the language setting.
• If recognition fails, for example because the engine files cannot be downloaded or the image is damaged, the page shows "Recognition failed. Please try again or use a clearer image".
• The first use needs an internet connection to download the recognition engine and language data from a third-party CDN (see the privacy section), so it fails when the network is unavailable. The page notes that Chinese plus English is roughly 10+ MB.
• Large images or images with a lot of text take longer. Recognition uses your own device's processing power, so it can be slow on low-end phones.

FAQ: Why does the result contain so many mistakes?

OCR estimates each character from its shape, so the clearer the picture the better the outcome. Typical causes of errors are low resolution, tiny text, a tilted image, a busy background, the wrong language setting, and look-alike characters in mixed text, such as the digit 0 and the letter O. Try the original file or a higher-resolution screenshot, crop away unrelated areas and select the right language. For numbers, amounts and names, always check each item by hand.

FAQ: Why is the first recognition so slow?

The first time, your browser has to download the recognition engine and the data for the chosen language before real recognition begins. On a slow connection this takes noticeable time, and the progress figure can stay unchanged for a while. Later runs with the same language are usually quicker, though that depends on how your browser caches files.

FAQ: Can it read handwriting?

The engine is aimed at printed text. Handwriting results are generally unreliable: neat handwriting may be partly recognised, and messy handwriting mostly will not. If handwriting is your main need, check the output carefully and do not depend on it.

FAQ: What is the difference between Chinese + English and English only?

Chinese + English loads both language datasets and suits mixed content, but it involves more data and can be slower. English only loads just the English data, which is faster for purely English images and often gives a cleaner result when there is no Chinese in the picture. Traditional Chinese, Japanese and Korean each use their own language data, so choose according to what the image contains.

Privacy and how files are processed

Your image is not uploaded to this site or to any server for recognition. The file is handed directly to tesseract.js running in your browser, and recognition happens on your device. To be fully transparent, though, on first use tesseract.js downloads its worker script, its recognition core and the data for the selected language from a public CDN (jsDelivr). That means a request to a third-party server, which can see ordinary request details such as your IP address but not your image. The recognised text is shown only on the page and is not saved; refreshing or closing the page clears it.