Apparently Xerox machines use JBIG2 for compression, an algorithm that creates a dictionary of image patches it considers similar. As long as the error generated by these patches is not too high, the machine reuses them instead of using the original image data. This also would explain why the error occurs when letters or numbers are scanned in moderate resolution, yet still readable for the human eye. When the letter size is close to the patch size of JBIG2, complete letters and even blocks of letters that look similar to the machine are replaced.
Stay up-to-date by subscribing to the Comments RSS Feed for this post.