Editing scanned PDFs with OCR

A scanned PDF is a picture of a page. There is no text inside it to select, search or edit, only pixels. blupdf recognises the text on every scanned page and rebuilds the page as a real document, then treats it exactly like a digital PDF: click a block, retype, download.

What counts as a scan

blupdf classifies each page when the file opens. A page is scanned when it carries no usable text layer: output from a scanner or a multifunction printer, a phone photo, a fax, or a PDF "printed" to an image. Mixed documents are handled page by page; a contract with two scanned signature pages and ten digital pages pays recognition for the two scanned pages only.

PNG, JPG and WebP images can be dropped into Studio directly. Each image is wrapped into a PDF page sized to the picture (long side up to 3,000 pixels) before recognition.

Recognition runs when you first open the page

A scanned page is recognised the first time you reach it in the editor; the page shows "Reading the text on this page…" for a few seconds. blupdf sends the page image to its text-recognition engine, a trusted third-party service built for documents, which returns every word with its position, its measured type size, whether it is bold, italic or handwritten, its color, and any checkboxes it finds. Typical time is two to four seconds per page. The result is stored with the document, so recognition runs once per page and costs 80 credits per page, charged after the page is recognised. If your balance runs out in the middle of a document, the page keeps its recognised text; only the charge waits.

The page is rebuilt as real text

Recognition alone would give you words floating over a picture. blupdf goes further and rebuilds the page: every line is drawn again as real glyphs in the closest library family, at the size the engine measured, in its weight, slant and color; horizontal and vertical rules are redrawn as vector lines; checkboxes become vector squares with their marks; logos, photos and stamps are cropped from the scan and placed where they were. The result is a clean, searchable document rather than a picture of one. The editor shows "Rebuilding this page…" while it is made.

The original scan is kept. Show original scan on the page toggles between the rebuilt page and the scan, so you can check a doubtful word against the source at any time.

Accuracy and what affects it

Document OCR is very accurate on clean office documents and noticeably less so on poor input:

• Resolution. 300 dpi is ideal. Below 150 dpi, small text and punctuation start to blur together.
• Skew and perspective. A page photographed at an angle stretches letters unevenly; scan flat when you can.
• Lighting and shadows. Phone photos with a shadow across the page lose characters in the dark band.
• Handwriting. Printed text is reliable. Handwritten entries are recognised and flagged as handwriting; check them line by line.
• Language. Printed text in most languages, Latin and non-Latin scripts alike. Mixed-language pages are fine.

Fonts on scans

A scan carries no font information, so the engine's style hints decide the family: sans, serif, monospace or handwriting, bold and italic, and the measured size. blupdf picks the closest family from its library and uses it for the rebuilt text and for anything you type. It blends with the page at reading size but is not the identical typeface, because that typeface is not in the file. The text toolbar shows the chosen family and lets you switch to another. Style and size are decided per row, so a line keeps one look instead of changing weight word by word.

Fixing recognition mistakes

Because the rebuilt page is just text in blocks, correcting a misread is an ordinary edit: click the block and fix the characters. Common misreads are 0/O, 1/l/I and rn/m. If a region was missed entirely, for example a rotated label, add a text block with the Text tool and type it. If a page could not be recognised at all, the editor says so and keeps the scan as it is.

Downloading

The PDF download contains the rebuilt pages, so the text is real, selectable and searchable, and screen readers can read it. DOCX gives you the recognised text as an editable Word file; JPG and PNG give you page images. See Pages & downloads.

Photos as input

Photos work like scans with the extra wrapping step. For best results, crop to the page, keep the camera parallel to the paper, and avoid flash glare. A photo of your signature is a different job: to place your signature on a document, use the signature tool described in Signing a PDF.