Image to Text (OCR)

The Image to Text tool extracts the readable text from a JPG, PNG, WEBP, GIF, or BMP image. It runs tesseract on the server, so the recognition works on any browser without a plugin, on phone or desktop, and without uploading anything besides the file you want OCR'd.

How to Use

1. Drop or pick your image. The drop zone accepts JPG, PNG, WEBP, GIF, and BMP up to 10 MB. HEIC, TIFF, and SVG are not supported directly — convert them to JPG or PNG first with the Image Converter.
2. Pick a language. Tesseract is multilingual; pick the language of the text in the image (defaults to English).
3. Extract. Click Extract text. The server runs tesseract and returns the recognized text inline. Copy it to your clipboard, or download it as a .txt file.

Supported Languages

The tool exposes a curated set of 14 languages. Tesseract ships with more, but the ones below cover the vast majority of requests:

| Code | Language | Code | Language | |------|----------|------|----------| | eng | English | chi_sim | Chinese (Simplified) | | spa | Spanish | chi_tra | Chinese (Traditional) | | fra | French | jpn | Japanese | | deu | German | kor | Korean | | ita | Italian | ara | Arabic | | por | Portuguese | hin | Hindi | | nld | Dutch | rus | Russian | | | | | |

If you need a language not on this list, the tesseract data files for many others are available — let us know and we'll add it.

Accuracy and Limitations

Tesseract is the most widely used open-source OCR engine and is generally very accurate on clean, high-contrast printed text. Specifically:

• Excellent (95%+ per word): screenshots, 300 DPI scans, black-on-white printed pages, clearly typed receipts, product packaging labels.
• Good (80-95%): phone photos of printed pages in good light, lower-resolution scans, multi-column layouts.
• Poor (below 80%, often unusable): low-resolution scans under 150 DPI, stylized or handwritten text, photos with shadows or perspective distortion, images with busy backgrounds behind the text.

Handwriting is not reliably recognized. If you need handwriting OCR, look for a dedicated service that uses a neural model (this tool doesn't).

The result page surfaces a mean per-word confidence from tesseract so you can judge how trustworthy the result is without leaving the page. When the image is too noisy for tesseract to recognize any words, the result will say "No text was recognised" rather than fabricate a guess.

Privacy

Your image is uploaded to the server for recognition. The image is staged to a temporary file, run through tesseract, and then deleted — both the input and tesseract's text/tsv sidecars are removed in a finally block, so the file does not persist between requests. Nothing about the image is logged.

Technical Details

• Engine: tesseract 5.x, invoked via proc_open with a 5-second wall-clock timeout per read.
• Input formats: JPG, PNG, WEBP, GIF, BMP. Up to 10 MB.
• Page segmentation: --psm 3 (fully automatic page segmentation, no OSD).
• Confidence: parsed from the TSV sidecar that tesseract emits when run with tsv config; the mean is computed only over word-level rows (level 5) and reported as an integer 0-100.
• Output format: by default, text/plain; charset=UTF-8 with the recognized text as the body. Add ?response=json to get a JSON envelope with text, language, sourceSize, textLength, and confidence (matches the freeconvertai.com/api-docs contract shape).
• Rate limit: 30 requests per minute per IP (lower than the resize/convert siblings because OCR is more expensive).
• Defense in depth: CSRF header required, language code is whitelisted (no untrusted strings reach the shell), escapeshellarg() on every arg, temp files cleaned up in finally.

FAQ

Does this work on scanned PDFs?

Not directly — this tool reads images. If you have a scanned PDF, use the PDF to Text tool first (which uses a different pipeline).

Can I OCR multiple images at once?

No — the tool processes one image per request. If you have a stack of images, run them one at a time. (A multi-file batch mode could come later; right now the focus is keeping the single-file case fast and reliable.)

Is there a character or word limit?

No hard limit on text length, but tesseract's accuracy depends on image quality and language, not text length. The tool will return however many words it can recognize from the image you uploaded.

Why is tesseract returning garbled text?

Almost always one of three causes: the image is too low resolution, the language is wrong, or the text is in a font tesseract hasn't been trained on. Try a higher-resolution scan, switch the language, and re-run. If it's still garbled, the image is probably a bad candidate for OCR in general.