Anyone can easily prove that the layers betrayed by Illustrator show intelligent manipulation of image and data, such as one layer entirely the background, another mostly printed text, another dates, another signatures, registrar stamps, etc.
This claim always made me laugh, because when you actually look at the objects in question and try to interpret them as productive Photoshop layers, you can't help thinking what sort of incompetent moron of an artist would try to use Photoshop that way. Or even how anyone
could, without going nuts.
No, the "layers" don't make any sense as a composed document in Photoshop. But the Birther claim is that the PDF exists as several different objects because the hypothetical artist composed the Ps layers that way and then consciously set up the PDF exporter to preserve those layers as separate objects. (Flattening is the default in most Ps versions.) Supposedly only the Birther pseudo-experts were smart enough to be able to see this allegedly bogus representation for what it was.
...virtually impossible to replicate with any compression program
Or exactly what you'd expect, as we shall see.
It's time we examined these nine
objects and try to determine what they are and why they're there.
"One layer entirely the background" is a
total lie. The object in question is the green safety paper,
and the ruling on the form,
and the signatures,
and random letters from the typewritten text -- such as the "R" in "BARACK" but not any of the other letters. In other words, a bunch of apparently unrelated elements.
An artist is going to create a layer for the crosshatched safety paper, a separate layer for the line art, and then another layer or layers for the signatures. He's also going to add an adjustment layer with a gradient mask to create the part of the image where the book gutter margin is darker.
Further, the contents of other objects have been "knocked out" of this background object so they are visible as little white outlines. This is necessary for how the PDF rendering engine works, but it is completely unwarranted in Ps. The Ps artist knows that higher layers will simply obscure the background according to the blend-mode settings.
The next object is
most of the text and some of the deep black graphic elements like arrows for the signature boxes. Keep in mind that a fair amount of other text appears in the "background-only" object. These elements are jagged and stark because the object is a bit mask. It can represent only the presence or absence of the graphics-state. During the importation, Ai assumes it will be black and converts it as such.
Why logically-related text in two different objects? Makes no sense in Ps, but it makes absolutely perfect sense in PDF optimization. PDF optimizers perform a high-pass filtration of the original image to detect where edges might occur, and where stark transitions from one color to another might be replaced by transitions that lie on pixel boundaries. This allows a portion of the graphic to replaced by a bitmap, which takes up much less space.
However, in the original image the "R" in BARACK was too fuzzy to trigger the high-pass filter. As such it got left as part of the "deep" pixel map so that the fuzzy edge could be accurately rendered using greater image depth. The deep image requires lots of bits per pixel so that it can accurately reproduce true graphical elements as well as "black" elements that nevertheless require some gray scale on their edges to aid interpretation. For example, long thin lines that don't precisely align to the raster, if rendered in 1-bit bitmask-type graphics, are unacceptably jaggy.
It is trivially easy to find and create examples of overly jagged items in optimized PDFs. In the magazine page I scanned and optimized, one photo was of the Ps Brushes palette. The optimizer extracted the hard-edged brushes and replaced them with a bit mask, while leaving the soft-edged brushes alone!
Throughout the document we see that the criteria for keeping an object on the deep background rather than the shallow text object is the requirement that the object in the original image have some grayscale properties, not its logical function in the purported Ps forgery.
Optimizer: 1
Artist: 0
The next two objects are the registrar's signature and the date under the seal, respectively. Again, these are bitmaps. But why weren't they included with the other line-art and text objects? Because they're geographically far removed from the rest of it.
In Ps the layers are the same dimension and depth. Hence you don't gain anything in terms of convenience or space by having separate layers for content that is geographically separate but logically and graphically related.
But in PDFs, image objects are no bigger than they need to be to express the element in question. Hence you gain considerable space savings by taking an image composed of geographically clustered elements, and breaking each element up into its own image. You don't have to store the empty space that way.
Optimizer: 2
Artist: 0
The next three objects are two date stamps and the first three letters of the word "None" -- i.e., omitting the "e" which is contained on the principal text and graphics layer.
The astute viewer will notice that while these too are 1-bit images, they are not full black. They are, in fact, a dark gray. The bitmap-conversion algorithm says, "Can this region be represented by a single color?" In most cases that color is black. Here the color is dark gray because while the edge-detection algorithm found an edge, and while the color-depth simplifier found a single color, it was not the
same color that had been used in the other single-bit image. Hence it can't be included with it, because the graphics-state color is different for that bit mask than for the other bitmap objects. Before rendering these bitmasks, the graphics state must be changed to the new color, so they have to be separate rasters.
And again because all the bitmask objects of this color are far separated, they are separate image objects, not all on the same layer.
In Ps a layer can contain any color compatible with the overall image's color model. There is no justification (and considerable disadvantage) to putting different colored objects necessarily on a different layer. But this is exactly how the optimizer renders single-color image elements using the minimum data storage.
Optimizer: 3
Artist: 0
The last two objects are alterations to the paper, one the imprint of the registrar's stamp and the other a set of discolorations at the top edge of he paper.
To understand why these are there, you need to understand the difference between component images and indexed images. Component images store each pixel as a tuple of the numbers that express the components of the pixel's desired color in the color model -- HSL, CYMK, RBG, YUV, etc. You may be familiar with component expressions of color such as
#0080ff, which represents RGB values for a color as six hexadecimal digits, two digits for each component. 00 means fully off, 0xff (255 in decimal) means full intensity, with lesser intensities represented by the intervening numbers.
In an indexed image, you predetermine a palette of colors you're going to use, and assign each a sequential number. The value stored for each pixel is the number, the index into that predetermined color table, which is a much smaller number than representing the pixel component values directly. The drawback is that your image must be composed only of the colors you've selected.
Converting a component image to an indexed image requires color-space quantization, that is, reducing the number of colors needed to visually represent an image. In virtually every case, you have to approximate the color in the source image using the closest entry in the color table. The "error" between that entry color and the actual color is accumulated and considered when deciding the colors of neighboring pixels. Since pixels are very small, the overall impression will be of roughly the correct colors. Artifacts of such a quantization and error-diffusion process appear all over the background object in the Obama PDF.
Of course we can simply look in the raw PDF data structures to see that the image has been quantized and indexed. But we can confirm that this process has taken place by magnifying the image. The green safety pattern contains "speckles" of error-diffused pixels in the white spaces. The pattern has been approximated in the PDF by using the minimal number of green colors.
8-bit quantization produces very compact image storage and allows up to 256 colors. You can choose what those colors are, but once chosen you are limited to them. How to choose? Well, by histographic analysis of the source image. This is a sophisticated statistical process, but the result is to determine the set of 256 colors that will best represent
all the colors in the image, using appropriate error diffusion.
Here a clever thing has happened. The optimizer has realized that if it eliminates certain regions of the background image that contain outlying pixel data, it can recover the colors it would need to represent them and add them to the greens and grays it needs to represent the bulk of the background and fuzzy line art. And since those outlyers are generally single-color aberrations, they fall under the auspices of changing the graphic state and applying a bit mask.
So here the quantization problem has been simplified by removing slightly outlying data and making it its own simplified object. No reason to do that in Ps since minor color and pattern variations -- especially to an established pattern, can happily live in a single Ps layer.
Optimizer: 4
Artist: 0
So the notion that an optimizer "wouldn't produce" this pattern of objects is simply as naive as it can be. This is
exactly the kind of way an optimizer would carve up a raw scan. The notion that these are the accidentally revealed layers of some Ps artist tasked with creating the forgery makes no sense whatsoever.
