Flaw Results In Xerox Scanners Making Costly Errors

Xerox has admitted that some of its scanners mix up numbers. It's a problem that could prove costly for business users.

Surprisingly, the problem isn't related to optical character recognition, which turns a scanned document into text; instead, the mixed-up numbers are right there on the scanned image.

David Kriesel, a computer scientist in Germany, discovered the problem when scanning construction documents. The documents Kriesel scanned were related to three rooms of different sizes.

However, after making the scan Kriesel found the document showed all three rooms as the same size -- a clear error and something that could have caused serious construction problems.

He then tested the scan three times on a different machine and discovered that not only did all three scans have at least one incorrect digit, but the error was different every time.

Compression Technique Creates Number Confusion

After highlighting the problem on his blog, Kriesel compared notes with other users and discovered the problem involves the way the scanner uses a compression technique employed when someone selects "normal" rather than "high" quality.

The technique, known as JBIG2, looks for repeated patterns in the image before carrying out a full scan. When it finds such patterns, it scans and stores the data for one area. For the other areas, it simply reproduces the original area.

Unfortunately, for small characters (under 10pt text), the scanner sometimes mistakenly assumes they are a repeated pattern. It scans one digit and then reproduces it in other places in the document, often incorrectly.

To make things worse, the false numbers are in the right size and typeface. This means there's no easy way to spot the mistake.

Xerox Says It Didn't Make Danger Clear

Kriesel says this could cause a wide range of problems, from misleading and confusing accountants to patients receiving the wrong dose of medicine.

Xerox says this is a known error and insists it has mentioned the issue in instruction manuals for 14 affected models. But the firm concedes it may not have made the problem clear enough.

It's now planning to issue a software update that will let users switch the feature off completely if they are worried about mix-ups.