Gratuit · Images
Image en texte (OCR).
Extrayez localement le texte de photos, captures d’écran et images numérisées grâce à l’OCR en 11 langues, dont l’arabe. Vérifiez, corrigez, copiez ou exportez en TXT et Word.
JPG, PNG, WebP, BMP or GIF (first frame) · up to 10 images, 15 MiB each, 60 MiB per run, 25 megapixels per image. Images stay in your browser; the OCR engine and selected language models load from this site (about 2–20 MB per model, cached after first use).
Printed text, screenshots and clear photos work best. Handwriting, decorative fonts, low resolution, blur, perspective and complex tables are unreliable. Recognition does not rebuild layout, tables or formatting. Always review the text before relying on it.
Comment utiliser cet outil: Image en texte (OCR)
Saisissez ou sélectionnez vos données. Réglez les options, générez le résultat puis vérifiez-le avant de le copier ou de le télécharger.
Limites et formats acceptés
Vos données sont traitées sur votre appareil.
Notes techniques en anglais
JPG, PNG, WebP, BMP or GIF first frame; up to 10 images, 15 MiB each, 60 MiB and 500,000 characters per run, 25 megapixels per image. Images are reduced to about 12 megapixels for recognition. Printed text only; handwriting, decorative fonts, blur, perspective and complex tables are unreliable. No layout, table or formatting reconstruction.
Recognition uses Tesseract LSTM models (tessdata_best) served from this site: English, French, Arabic, Spanish, German, Portuguese, Italian, Dutch, Hindi, simplified Chinese and Japanese, up to three per run. Images are decoded with their EXIF orientation, optionally rotated, enlarged when text is small and cleaned up: automatic mode converts to grayscale, stretches contrast and inverts light text on dark backgrounds. Layout modes cover automatic page segmentation, a single column, a single block or a single line; automatic mode retries a single block when nothing is found. Confidence is reported per image and words below 80% are highlighted, but confidence is not an accuracy guarantee. Text is editable before copying or export; Word export places one image per page and does not include the images. Generated example checks are not a general accuracy measurement. Images stay in the browser.
Paste a dark-mode screenshot with small text: automatic cleanup inverts and enlarges it before recognition, then copy the corrected text.