NBC has replied to your comment (excerpted to exclude snark):
http://nativeborncitizen.wordpress.com/2013/08/11/mmaschin/
MMaschin: NBCs argument is easily shown to not be valid, because if that was the case, then the perfect place for this to have occurred is in the security background layer.
NBC: The security background layer is recognized as a colored background and is almost completely separated as such. Since it is compressed with JPEG, there are no JBIG2 artifacts.
MMaschin: That IS a contiguous, repeating pattern. Are we to believe that a computer process found two hand draw, ink on paper, and photocopied boxes similar enough to replace one with a copy of the other, but it did not find a pattern in the security pattern background? If what NBC says is true, then the compression occurred AFTER the image layers were separated, because if you look at the image as a whole the boxes are very different when you considered the content inside the boxes.
NBC: That is the whole point of Mixed Raster Compression, the colored background is compressed differently than the foreground ‘text’.
MMaschin: This means that even analyzing the background layer by itself, this compression algorithm was incapable of finding a repeatable pattern.
NBC: MRC does not look for repeatable patterns, it only happens when the foreground bitmaps are compressed.
MMaschin: What NBC is saying is not taking into consideration that this must be a systematic process, and not a logical one. If you say ‘this was done here’, it must hold true for the entire document, you can’t cherrypick
NBC: I am consistent in my hypothesis.
The image is scanned, and segmented into a background image and multiple foreground images. The background is JPEG compressed with relatively high quality reduction and then subsampled to half the resolution. The foreground bitmaps are JBIG2 compressed (which encodes repeating patterns).