lberki / qr Goto Github PK

This causes problems with large QR codes where modules are small enough that e.g .barrel distortion becomes significant.

This is why hugeflat cannot be scanned at the moment.

Parsing a QR code from a 640x480 image takes ~250 milliseconds on a beefy laptop.

This is clearly wrong and I have no idea where time is spent.

It compares every pixel to its local neighborhood of a constant size. Thus, if the module size is larger than that, it will fail (because a module is full-black or full-white).

Was surprising to learn that a perfectly aligned, crisp, black-and-white QR code image cannot be scanned for this reason.

One solution would be to run binarize after the finder patterns were found and use only the pixels in the run being checked for the black/and white decision during pattern finding. However, it would make things even slower.

Shouldn't be too hard to do something about it. The code is in geom.js in fitLine.