OCR PDF

OCR turns a scanned PDF into a searchable one. This tool reads every page with optical character recognition and writes an invisible text layer over the original image, so the document looks untouched but its words can be searched, selected, and copied. All of it happens inside your browser.

Why a scanned PDF is so frustrating to work with

A scanner, photocopier, or phone scanning app produces a PDF made of flat images. It looks like a document and behaves like a photograph, which causes four specific problems: Ctrl+F finds nothing, so you cannot locate an invoice number or a name; text cannot be selected or copied into a spreadsheet or email; screen readers cannot read it, which makes the file inaccessible; anddesktop search cannot index it, so it is invisible to Windows Search and macOS Spotlight. A folder of scanned contracts is effectively unsearchable until you fix it.

How the invisible text layer works

Each page is rendered to a canvas at double scale, which gives the recogniser more pixels to work with than the page's nominal size. Tesseract then returns not just the words but a bounding box for each one. The tool converts those coordinates back into PDF space, accounting for the fact that PDF measures upward from the bottom of the page rather than down from the top, and draws each word at zero opacity in its correct position over the untouched page image. The visual result is byte-for-byte what you scanned; the functional result is a document your reader treats as text. This is the same technique used by commercial OCR suites.

ToolJiffy compared with cloud OCR and desktop suites

Getting the most accurate text layer

1. Scan at 300 DPI. Below about 200 DPI, small print starts to break down whatever the engine.
2. Set the language before processing. This single choice affects accuracy more than anything else.
3. Straighten the pages. Skewed scans lose accuracy quickly; most scanner software can deskew automatically.
4. Scan in greyscale or black and white for plain text documents, which removes colour noise from the page.
5. Check a few pages afterwards by searching for a word you know appears, rather than assuming every page worked.

Privacy for the documents people actually scan

The files that need OCR are rarely trivial. They are tax returns, medical records, bank statements, signed agreements, identity documents, and case files. Sending those to a cloud OCR endpoint means a third party holds a full copy of both the scan and every word extracted from it, and you inherit their retention policy and breach history. Local processing removes the question: there is no upload, no temporary server-side file, and nothing to chase for deletion afterwards.

Where OCR stops, and what to use instead

A text layer makes a document searchable; it does not rebuild its structure. If you need the numbers out of a scanned table, PDF to Excel reconstructs rows and columns rather than a flat run of words. For a single image rather than a PDF, Image to Text is the quicker path and hands you editable text directly. If you only want the pages as pictures,PDF to JPG exports them, and if the finished searchable file is too large to email, the PDF compressor will bring it down.