A QR code label gets stained, or a small patch in the middle is covered. Sometimes a phone still reads it successfully. That seems surprising. If some of the black and white squares are missing, shouldn't some of the information be missing too? The answer involves two separate jobs: finding the code in an image, and recovering the information inside it.

The history helps explain why both jobs matter. According to DENSO WAVE, users of conventional barcodes wanted to store more information, including Japanese kanji and kana. The development team wanted a new code that could hold more data and be read quickly. The QR code was released in 1994. Its name stands for Quick Response, reflecting that emphasis on speed. The automotive industry subsequently adopted it for electronic kanban, supporting work from production to shipping.

Imagine an ordinary QR code as a black and white checkerboard. At three of its corners sit larger square targets. These are position detection patterns. They help the reader locate the code and determine its orientation. Rotating the label does not necessarily prevent reading, because the reader has recognizable landmarks to work from.

There is a distinctive pattern inside each target. Across its center, the widths of the alternating black and white regions follow the ratio one, one, three, one, one. DENSO WAVE's history describes how the developers studied printed material to choose a ratio unlikely to be confused with surrounding graphics. You do not need to memorize the ratio. The useful idea is that a code carries recognizable structure to tell the reader where the information is.

Finding that structure is only the first job. Recovering damaged information requires error correction. When a QR code is generated, additional correction information is encoded alongside the original data. DENSO WAVE identifies the method as Reed-Solomon coding. Within its correction capacity, it allows the reader to restore data affected by dirt or damage.

A simple arithmetic analogy can make redundancy easier to understand. If you know two numbers and their sum, losing one number may still leave enough information to recover it by subtraction. A QR code uses much more sophisticated mathematics. It does not simply store a second complete copy of the message. The shared principle is that additional relationships between pieces of information can make recovery possible.

There are four correction levels. The official table gives approximate restoration rates of seven percent for level L, fifteen percent for M, twenty-five percent for Q, and thirty percent for H. The crucial detail is what these percentages measure. DENSO WAVE defines them relative to the total number of codewords. A codeword is an eight-bit unit of encoded information. It is not one visible square, and it is not a unit of image area.

Level H therefore does not promise that covering any thirty percent of a code's surface will leave it readable. Which structures are affected, how damage translates into errors in the encoded information, and whether the reader obtains a usable image all matter. The official feature description explicitly warns that restoration may not be fully achieved, depending on the dirt or damage. One decorated code scanning successfully is evidence about that particular code under those particular conditions.

Higher correction also has a cost. More correction information must fit into the symbol. For the same message, that can require a larger symbol version. Keeping the version fixed instead reduces the amount of original data it can hold. Squeezing a denser symbol into the same printed space makes individual squares smaller, which can introduce camera and printing limits. Stronger correction cannot solve every image problem.

The space outside the code matters too. DENSO WAVE specifies a clear margin, called a quiet zone, four modules wide on every side of an ordinary QR code. A module is one small square. This blank border should contain no other printing. Keeping that border, the position patterns, clear contrast, and an appropriate module size supports reliable reading.

For a label that will not scan, cleaning its surface, reducing glare, and obtaining a sharp image are practical first steps. For a label you produce, preserve the structure and margin, then test the actual size, material, and reading devices. The official FAQ warns that adding illustrations or altering a code can slow reading or make even a small missing or smeared part cause failure.

A partly covered QR code sometimes works because its design includes both location cues and error correction. That reserve helps it survive imperfect conditions. It does not guarantee arbitrary damage will be harmless.
