Above the amount field in the transaction form sits a small camera card. Tap it, take a photo of a receipt, and three fields get pre-filled: the total, the date, and the merchant name.
The OCR engine runs on your phone. The receipt image is processed in the app's sandbox and never sent to a remote service, so card numbers, merchant details, and anyone you were shopping with stay between you and your camera roll.
OCR fills fields, it does not write them. Every pre-filled value is editable before you save - if the engine misread a smudged total, you correct it. The same flow exists on the bank card form: tap the OCR card there to read a card number from a photo. In both cases, the photo is dropped from memory after recognition; Money Tracker stores the recognised text, not the pixels.