Nine pixels per square

Every square in a QR code is drawn here as a 3×3 block. Only the middle pixel is held to the value the code needs — the other eight are free to carry the picture. Decoders sample each square near its centre, so the part they read stays true while everything around it becomes image.

The landmarks stay untouched

The three corner squares, the dashed timing lines running between them, and the smaller alignment blocks are drawn at full strength, with no image in them at all. A scanner needs those to spot the code, square up the perspective, and lay its grid over the rest before it reads a single bit.

Error correction covers the rest

QR codes carry Reed–Solomon redundancy, in four levels. This uses the highest, level H, where roughly 30% of the code can be unreadable and it still decodes. That margin absorbs the squares where a dark passage of the photograph pulls the reading the wrong way.