What image quality is needed to scan an MRZ successfully?

The MRZ's fixed-pitch OCR-B font is genuinely forgiving compared to general text recognition, but there are still practical minimums for image quality that matter in real-world scanning. Resolution needs to be high enough that individual characters in the MRZ are clearly distinguishable rather than blurring into each other.

On a typical smartphone camera, this usually isn't a limiting factor as long as the phone is reasonably modern and held at an appropriate distance, close enough that the MRZ occupies a meaningful portion of the frame rather than being a tiny sliver of a wide shot.

Focus matters more than raw resolution in practice. A high-megapixel camera pointed at a document that's slightly out of focus will still produce a poor result, while a lower-resolution camera with a sharp, in-focus shot often performs better.

This is part of why good scanning apps use autofocus and sometimes require a brief pause before capture to let the camera lock focus properly.

Lighting is the other major factor. Even, diffuse light works best.

Harsh direct light can cause glare off the document's surface, particularly if it has a laminated or glossy finish, which can obscure characters entirely in the glare's path. Very dim lighting, on the other hand, increases noise and reduces contrast between the text and background, making characters harder to distinguish.

Neither extreme is ideal, and most identity documents have at least some reflective coating that makes glare a genuinely common practical problem, not just a theoretical one.

Angle and framing round out the practical requirements. The MRZ should be roughly level and fully within the frame, without being cut off at the edges, since a partially captured MRZ obviously can't be fully read.

Motion blur, from a shaky hand or a document that's not held steady, is another common real-world issue distinct from focus problems.

ScanDoc's mobile SDKs handle much of this through real-time capture guidance, detecting and flagging glare, blur, or poor framing before an image is submitted for processing, which meaningfully reduces the number of failed or low-confidence scans compared to a simple "take a photo" flow with no in-the-moment feedback.