How do barcodes work?

From a number to bars

Every digit in a UPC or EAN takes seven columns of equal width, called modules, and is drawn as two bars and two spaces. Each bar or space is one, two, three or four modules wide, and the four widths always add up to seven. GS1 defines three sets of patterns for the digits, called A, B and C. A zero in set A, for example, is a space three wide, a bar two wide, a space one wide and a bar one wide.

Three guard patterns frame the digits. A guard of bar, space, bar sits at each end and a five column guard in the middle. They tell the scanner where the symbol starts, where it ends and how wide a module is. Add them up and a UPC-A is 95 modules across, which IBM's history page describes as 30 black lines and 29 blank ones. The UPC generator draws your number and lights up the columns for any digit you pick. What the digits mean is in what a UPC code is.

Reading either way up

The six digits on the left of a UPC-A come from set A, where each pattern has an odd number of dark modules. The six on the right come from set C, which is the mirror image of set B, and have an even number. A scanner that passes a guard and then meets odd patterns knows it is reading left to right. Even patterns mean it is reading backwards and should reverse the result. That is why a cashier can swipe a product in either direction, and why GS1 calls the EAN and UPC family omnidirectional.

What the scanner sees

A scanner sees changes in reflected light along a line rather than a picture of bars. A pen reader is dragged across the bars and a photodiode turns the bright spaces and dark bars into a waveform. A laser scanner sweeps its beam with a spinning mirror, between 200 and 1,200 times a second, and a photodetector reads the reflection. A CCD reader has a row of hundreds of light sensors and measures the light already falling on the code. A camera scanner, which GS1 calls an imager, takes a picture and finds the bars in it.

The difference matters for what each can read. Zebra says a laser scanner reads printed linear codes well, even poor ones. Zebra adds that it cannot read a code on a phone screen and cannot read a 2D code. A camera imager reads printed and on-screen codes, both linear and 2D, from any angle. GS1 assumes a retail scanner reads from about 100 mm away. Wikipedia gives 0.33 mm as the most common narrow bar width, with some scanners reading bars as thin as 0.075 mm.

Whatever the hardware, the decoder measures the distance from the edge of one bar to the edge of the next similar one. It compares those distances with the width of a whole digit and picks the digit whose pattern fits. GS1 publishes that reference method so that any maker's scanner reads any maker's symbol.

Why red bars disappear

Scanning lasers shine red light, and GS1 grades barcode print quality with light at 660 nanometers for the same reason. Under red light, red ink reflects almost as much as white paper, so GS1 US says red or reddish bars are invisible to a laser. A reddish background is acceptable for the opposite reason. Black bars on a white background remain the safe choice, printed in a single ink.

Clear space, height and size

The scanner needs empty space on both sides to find the first and last bars. GS1 asks for 9 modules of clear space on each side of a UPC-A. That is 2.97 mm at the standard module width of 0.330 mm. An EAN-13 needs 11 on the left and 7 on the right. The whole symbol may be printed from 80 percent to 200 percent of that standard, a module of 0.264 mm to 0.660 mm.

Height matters as much as width. GS1 sets the standard bar height of a UPC-A at 22.85 mm. It says cutting the bars shorter, called truncation, destroys omnidirectional scanning. That is because a short symbol only reads when the item is turned to the right angle. GS1 also says never to put a barcode or its clear space over a seam, fold, flap, perforation or tight curve.

The digits printed under the bars are for people, not scanners. GS1 says they are not meant to be machine read. Wikipedia credits Murray Eden at IBM with adding them as the fallback for when the reader fails.

What the scanner sends

Once the decoder has the digits it recomputes the last one, the check digit, from the others. GS1 says the reader carries out that verification automatically. Seagull Scientific's BarTender help describes a failed check as a silent miss: the scanner sends nothing and the cashier scans again. What that sum catches is in what a check digit is.

The output is the number, nothing more. Most scanners connect as a keyboard, so the digits appear wherever the cursor is. Some can add a three character prefix that names the symbol type, such as ]E0 for a UPC or EAN. The price, the product name and the stock count live in the retailer's system and are looked up by that number. A barcode holds no price, and it holds no circuit either, so it cannot trip a store's security gate. The guide on store alarms explains what does.

Grading a printed barcode

Retailers judge a barcode with a verifier. It is a measuring instrument that looks at the symbol the way a scanner does and reports a grade from 4 down to 0. GS1 asks for at least 1.5 for retail symbols, measured with a 0.15 mm aperture under 660 nm light, which it writes as 1.5/06/660. The grade is the average of the grades of several scan lines across the height of the symbol. A verifier does not check that the printed digits match the bars, so proofread those yourself.

Reading one by eye

The printed line under a UPC-A shows all 12 digits, with the first standing left of the bars and the last standing right of them. The first 6 to 10 digits are the company prefix, the next 1 to 5 the item reference, and the last the check digit. A UPC-E, the short symbol on small packs, shows 8 digits. Where the UPC code is on a product covers where to find it on bags, cans and books. The check digit tool tests a number you have read off a label.

The same idea extended to a grid is in what a 2D barcode is. The people who worked all this out are in who invented the barcode.