Reading scanned and photographed statements with on device OCR
Jun 30, 2025
Scans and phone photos used to be a dead end. On device text recognition reads the page, corrects rotation, and flags the lines it was unsure about.
A photocopy, a fax, or a photo taken on a phone has no text layer to read, only pixels. Text recognition that runs on your device looks at those pixels and works out the characters, so a scanned statement becomes rows and columns like any other.
Rotation is corrected automatically, because phone photos are rarely square to the page. Lines the reader was unsure about are flagged in the preview rather than guessed at, which is the honest way to handle a smudged number.
Because the recognition runs locally, the scan never leaves the browser either. The same privacy that applies to a text PDF applies to a photo of a statement.
