QR Code Generator

A QR code is your data plus the arithmetic that lets a camera recover it from a bad photograph. There are four steps.

First the text becomes bits. This encoder uses byte mode, which takes the UTF-8 bytes of whatever you type, so it handles any language and any emoji at a cost of one to four bytes per character. The spec also defines a numeric mode and an alphanumeric mode that pack digits and uppercase letters more densely, which is why a code holding only numbers can be smaller than one holding the same count of mixed characters.

Then comes the error correction, which is most of what a QR code is. Reed-Solomon parity codewords are computed over GF(256) and appended to the data. GF(256) is a finite field with 256 elements where multiplication is addition of logarithms. At level L about 7% of the symbol can be destroyed and still recovered; at level H about 30% can. That is the whole reason you can print a logo over the middle of a QR code and have it still scan, and it is why the levels cost capacity: at level H a version 1 symbol holds 7 bytes where at level L it holds 17.

The data and parity are then split into blocks and interleaved before being laid into the grid. If a block's codewords sat together in the image, a coffee ring over that area would destroy one block completely and Reed-Solomon can only fix a bounded number of errors per block. Interleaved, the same ring damages every block a little, and every block stays inside its recovery budget.

Finally the grid is masked. Large blank areas and accidental finder-like patterns confuse a scanner, so the spec defines eight mask patterns, requires the encoder to score all eight against four penalty rules, and requires it to pick the lowest. The rules penalize runs of five or more, blocks of 2x2, patterns that look like a finder, and deviation from half-dark.

The quiet zone, four modules of blank space on every side, is part of the specification; codes printed hard against a border are the most common scanning failure.