Recognize Text via OCR

OCR stands for optical character recognition. In a PDF it means reading the page as an image, detecting the shape of each printed character and converting those shapes into actual text data. A scanned PDF stores only pixels, so nothing is searchable until recognition runs. Afterwards you have text you can copy, download or place back over the scan as an invisible layer.

Upload the scan, pick the document language and run recognition. The engine maps every recognised word back to its position on the page, and a searchable PDF keeps the original page image with that text written invisibly underneath. Visually the document looks identical, but search, text selection and indexing now work because real characters exist in the file.

PDF files and single images in JPG, PNG, GIF or WebP format. PDFs are rendered page by page before recognition, while images are fed straight to the engine, which is why an image to text job usually finishes quicker than a long scanned report. Photos of printed pages work as long as the text is in focus and the page is not strongly tilted.

Each language uses its own trained character set and dictionary. That data decides how ambiguous shapes are resolved, for example separating an accented character from a plain one, or distinguishing similar Cyrillic and Latin letters. Choosing the wrong language forces the engine to guess with the wrong reference, which produces garbled words. Nineteen languages are available, including Japanese, Korean, Chinese, Arabic, Hindi and Thai.

Yes. Enter page numbers and ranges such as "1, 3, 5-7" and only those pages are rendered and recognised, which cuts processing time on long documents. Leaving the field empty covers the whole file. If you would rather work with a smaller document from the start, split it first with Extract Pages .

A converter reads character data that already exists inside the PDF and rebuilds it as paragraphs, fonts and tables in a DOCX file. Recognition creates that character data from scratch by interpreting pixels, because a scan contains none. So on a scanned file PDF to Word alone may return empty pages or images, while text recognition returns the wording without the original formatting.

The downloadable output is plain text, so fonts, colours and table borders are not reproduced. Reading order follows the blocks the engine detected, which works well for single column pages and can interleave text on complex multi column layouts. If you need the visual layout preserved, use the searchable PDF output instead, since it leaves the page image untouched.

Almost always the language is wrong or the scan quality is too low. Recognition depends on clean edges between ink and background, so blurry photos, heavy JPEG compression, shadows and skewed pages break character shapes apart. Rescan at around 300 DPI, keep the page flat and evenly lit, confirm the language, then run the file again.

No. The character models are trained on printed type, where letters have consistent shapes and spacing. Handwriting varies per writer and joins letters together, so the segmentation step cannot isolate reliable glyphs. Handwritten notes, signatures and decorative script fonts will either be skipped or returned as nonsense, while printed forms with handwritten entries still yield the printed parts.

Recognition needs real computing power, so the document is uploaded and processed on the server rather than in your browser. The file is used for that job only and is not stored permanently. You can also clear the result yourself with the "Remove" button once you have downloaded the text file or copied it to the clipboard.

No. The uploaded document is read, not rewritten, and the recognised text is delivered separately as a .txt download or clipboard copy. If a searchable PDF is generated, it is a new file in which the page images are identical and only an invisible text layer has been added. Your local copy on disk stays exactly as it was.