Not to mention when the White House published Obama's birth certificate as a PDF. I remember being able to open it and turn the different layers off and on.
You can still open it with Illustrator if you want to see: https://obamawhitehouse.archives.gov/sites/default/files/rss...
So I just looked into this, and it's specifically Mixed Raster Content pipeline (ISO/IEC 16485) used in lots of different scanners. There's no need to find which specific software generated it because it's used by lots of them.
It's a technique used to attempt to isolate font characters of the same size and style as separate layers before OCR-ing to make OCR more accurate.
ABBYY FineReader, for example, is mentioned as producing the exact same type of results. But there's no guarantee that was the actual software because lots of scanning software does it -- it's a general technique. Plus it won't even be deterministically reproducible if it was e.g. scanned and OCR'd at higher resolution and then saved at a lower resolution, as is generally considered best practice for maximizing accuracy while keeping file sizes lower.
https://www.obamaconspiracy.org/2013/01/heres-the-birth-cert...
So this is very much a nothingburger. It's not an "alternative theory", it's a complete and total explanation.
Scan (600 dpi) > MRC (600 dpi) > OCR (600 dpi) > Downsample (150 dpi) > Save to PDF (150 dpi)
The image is saved in raster format at 150 dpi. That's the document, but not at the original scanning resolution. If you performed MRC and OCR at the 150 dpi level, you'd get different/worse results than were originally gotten at 600 dpi. Which is why you always OCR before downsampling, and you downsample for smaller files.
This isn't changing anything about the Snopes article. It just explains why if you run MRC/OCR at the PDF resolution, you won't deterministically reproduce it because it's not the resolution it was originally run at.
You do understand that this OCR is only for being able to search and highlight text? It's not changing what's displayed. That's still the pixels.
https://obamawhitehouse.archives.gov/sites/default/files/rss...
If you don't see jaggy pixel edges to the letters and form elements, what do you see?
I genuinely don't know what you're saying.
> Like, do you think that I think that scanners have come to operate on pure platonic forms and no longer use the concept of pixels? That would be really cool, wouldn't it.
Yes, because that is absolutely a thing. That's what Adobe ClearScan does, converting pixels to smooth vector outlines. Zoom in, and zero pixels in OCR'd text. That's not the case in this file though.
> I have yet to see someone show the original scribbles or ink marks that these OCR layers were generated based on. That's what I meant by "destructive".
I still genuinely don't know what you mean. The original scribbles and ink marks are a physical piece of paper. The MRC layers are generated from the scan and don't destroy anything, they only separate. The resulting layered bitmap is identical, pixel-for-pixel. My best interpretation of what you're saying is you want a higher-resolution scan? But why? Again, nothing "destructive" has happened except maybe reducing the resolution. But "destructive" is not a word people usually use for that.
> so you might want to just cut your losses and stop trying to educate me and let me be uneducated in this matter
I can't tell if you're being sarcastic or not. I am genuinely happy to help you understand, but if you really don't want to then obviously I won't spend any more time replying. But if you're going to publicly throw suspicion on the validity of the birth certificate, I feel it's important to correct the record here on HN simply for other people who might read this exchange.
> Now, it seems like you were saying that the text layers are just the pixels from the page moved up to a new layer. I said that I think that's surprising.
That's indeed all it is. It may be surprising, but that's how it works. Absolutely nothing about the pixels are changed as part of the process of that text layer separation. Later when saving the final PDF there's normal lossy compression, same as any JPEG, but the layer process actually preserves the text edges better than JPEG.
So I do hope you're satisfied now that everything about this is just normal image processing, and that nothing has been destroyed, there's no "missing evidence" or anything. And that the whole idea that the layers imply some kind of manipulation or forgery is false. What you're seeing is just the scan itself, saved (presumably) at a lower resolution and with the normal image compression scanners produce. The layer separation process is completely non-destructive and doesn't manipulate pixels at all.
This is easily explained by the character separation software being not 100% accurate.
It's not at all explained if someone is fraudulently adding text. Why would someone put half of the character in 1 layer and half of the character in a different layer?