Scanning Documents for Text Online: Where Does the Image of Your ID Actually Go?
Published 9 July 2026 · 5 min read
OCR — pulling actual, selectable text out of a photo or scanned image — gets used for all kinds of ordinary things: digitizing a stack of paper receipts for expenses, pulling text off a whiteboard photo after a meeting, or making an old scanned contract searchable. It also gets used, quietly and often, for a lot less ordinary things: reading the text off a driver's license or passport photo, extracting numbers from a scanned cheque, or converting a photographed medical form into text.
The tool doesn't know or care which category your image falls into. It just needs the actual picture, in full detail, in order to recognize what's written on it.
Why OCR needs more, not less, access to the image
Text recognition is genuinely demanding work. Recognizing handwriting, faded print, or a photo taken at a slight angle requires analyzing the image pixel by pixel, often multiple times, to correctly identify each character. There's no way to do that with a blurry preview or a stripped-down version of the file — the tool needs the real image, at real resolution, to have any hope of reading it accurately. Which means whatever's in that photo is exactly what gets uploaded, in the clearest form you have of it.
How this works locally instead
FormatDog's OCR tool runs Tesseract.js, an open-source text recognition engine, entirely inside your browser via WebAssembly. Your photos or scans are analyzed on your own device, using your device's own processing power, and the extracted text is generated locally. The images you're scanning, ID photos or otherwise, are never uploaded anywhere to make that recognition happen.
Because it's your device doing the actual recognition work rather than a shared server, you can process a batch of scans back to back without waiting in a queue behind everyone else also using the tool that moment.
How to check this for yourself
Open your browser's developer tools (F12), go to the Network tab, then upload an image and run the text extraction. If the tool is uploading your image to a server, you'll see that show up as outgoing network activity. If nothing appears carrying image data, the recognition happened entirely on your device.
Being reasonable about it
Scanning a recipe card or a whiteboard photo carries essentially no risk regardless of which tool handles it. The real consideration is the other, much more common use case for OCR that people don't always think about as sensitive — ID documents, financial paperwork, medical forms — where the image itself is the sensitive part, not just the text that comes out of it.