Butterdezillion – Some good questions

From the FreeRepublic we receive some good feedback from a poster named Butterdezilion.

Butterdezilion: If the Xerox machine is substituting exact replicas every time a certain “blob” (such as a box) appears, then that should happen with every box, every letter, etc. If the Xerox is switching 6’s for 8’s then where are those numbers switched around in the White House PDF?

A good question but as I have shown and found out, the Mixed Raster Compression is all but exact as it appears to be extremely sensitive to small variations. I have seen examples with anywhere from 4 to 17 foreground images. The same for JBIG2, it is based on how similar the two blobs, such as a box are, and in the samples I have, I have found JBIG2 to fail to capture the boxes, but it does capture other letters.

Butterdezilion If NBC scanned in the White House document which is a print-out of a document that ALREADY had these effects within it, then the real test would be whether the scan he comes up with is DIFFERENT than the White House PDF, not whether it is the same.

Exactly, it is not perfectly the same, or otherwise there would be reasons to doubt my findings. But the scan shows evidence of all the artifacts I mentioned, other than the halo effect which was of course already present in my ‘original’

Butterdezilion: If scanning the White House document using the Xerox machine under these conditions results in a PDF that hasn’t been manipulated by the Xerox (different than the document that was scanned in), it would actually CONTRADICT the theory that the Xerox made the anomalies in the actual content of the White House image.

The Xerox scanner does not ‘know’ that the document it scans used to have layers in it, and therefor it reproduces as best as it can the Mixed Raster Compression it is instructed to use by the default workflow. This results in similar but not perfect matches.

Even having access to the original Long Form Birth Certificate would not fully replicate the PDF provided by the WH, however I have now shown that the following features are captured

  1. Separation into a jpeg background and multiple monochrome foregrounds
  2. Alignment of two boundaries with 8×8 bit offsets
  3. Alignment of two boundaries with internal object
  4. Downsampling the background to 150 DPI
  5. Downsampling the foregrounds to 300 DPI
  6. Saving the JPEG with a quality factor of 47.48% with specific Quantization Matrix
  7. Embedding a YCbCr comment into the generated JPEG
  8. JBIG2 compression
  9. Preview created clipping mask
  10. Images are all in landscape direction requiring 90 degree ccw rotations
  11. The JPEG contains the same JPEG comment YCbCr
  12. The JPEG contains the same quantization matrix
  13. Full or almost full separation of the date and signature stamp
  14. Speckled foreground bitmap

What I have yet to explain includes:

Halos Although I believe the workflow provides us with plenty of hints there and preliminary experiments are looking promising. It’s just that I am working through the list somewhat methodically, often repeating experiments based on the feedback from others like RC, Goregan, Vicklund and our friend Hermitian.

Is that it? Just the Halos? Did I miss something? Surely there are more artifacts that were claimed to be evidence of a forgery and which can be explained through the workflow I propose?

4 thoughts on “Butterdezillion – Some good questions

  1. thecodont asks

    I am not an expert on this, but would Xerox be using a grayscale (8 bit) to black and white (2-bit) conversion for scans of these text images? Grayscale, of course, provides more information for further processing. You would have to have multiple scans of 8-bit to 2-bit of the same test image and then examine the results under a magnifying glass.

    Nope, a monochrome bitmap requires 1 bit (on/off) with 2 bits you can create 4 colors (00, 01, 10, 11). As to looking under a magnifying glass, remember that these are digital documents so we can use better tools. The reason to use B&W is that the image can be stored in 8 times as little information. It’s all about compression here.

    Good questions. Finding the identical characters is not much fun and I wish I had a tool to do so. But alas.

  2. Butterdezilion: And besides all that, the background pattern is not the one on the security paper that the HDOH actually uses. I took a photo of a valid COLB issued by the HDOH 2 weeks after Obama’s was issued, and the cross-hatch pattern does not match what is on the background of Obama’s. It matches the background that is on Virginia Sunahara’s HDOH-issued birth and death certificates.

    When they made the background layer they screwed it up. It takes no computer skill at all to see it when you know what you’re looking for.

    So you compare a COLB with a Long Form Birth Certificate security paper. Of course this is of little interest to my experiments which look at the PDF artifacts. It would be interesting to learn more about your findings. The background certainly matches up with the security paper used for President Obama’s COLB

    Your claim is interesting but hardly establishes a forgery unless you can show that the document was not created by the DOH of HI and all the evidence points to the contrary.

  3. One of Butter’s most persistent and bemusing conceptual failures lays in her complete reversal of the relationship between theories and data. Instead of using data to test her theories, she uses the theory as the filter by which to accept or discard data. The best example is in her hopeless confusion over the certificate numbering system. She theorized (without any evidence) that they would be numbered in date of birth, time of birth order and then declared the President’s number fraudulent by comparing it to those of the Nordyke twins. But as more and more numbers from August 1961 became available, none of them conformed to her original numbering theory. By logic, this should have resulted in discarding (or at least greatly modifying) the theory. Instead, Butter stuck to her theory and declared essentially every known certificate number to be a fake.

    Likewise here, she theorizes (again, without evidence) that “If the Xerox machine is substituting exact replicas every time a certain “blob” (such as a box) appears, then that should happen with every box, every letter, etc.” In point of fact, it is a trivial exercise to determine that this is simply not true. It might be a coherent theory, it is simply a false one as proven by the data. JBIG2 compression is not that consistent, and the reason it is deliberately not that consistent is no better demonstrated than by the recent examples of an algorithm so rigid that it actually changed numeric values on scanned documents.

    NBC’s process here has been rigorously scientific. He has tested his theory with the data, not the other way around. That is the difference between a scientific approach and that of the crank.

  4. Likewise here, she theorizes (again, without evidence) that “If the Xerox machine is substituting exact replicas every time a certain “blob” (such as a box) appears, then that should happen with every box, every letter, etc.” In point of fact, it is a trivial exercise to determine that this is simply not true.

    That’s the power of the scientific method is that it allows experiments to determine how correct the hypothesis is. And I am in no rush towards my conclusions, I report, step by step, review, check and onwards.

    It’s a deliberate and sometimes slow process.

Comments are closed.