The Tree

[davidn]

I suppose I might as well throw this up here as well, just on the off-chance that any developer-type friends have seen the image manipulation symptoms before. We're doing something that involves converting PDF pages to images on the fly using the PDFBox library, but certain images that look perfectly all right embedded in the PDF come out a bit... wonky. For example:



I think that I'm quite close to identifying what's wrong, and after analyzing the differences in how two libraries handle the colour models and spaces of PNG-encoded images, I have managed to produce this.



Never let it be said that my work doesn't have a noticeable effect.

Comments

[ravenworks.livejournal.com]

See, that first image is PRECISELY what I'm talking about when I say that there should be demoscene competitions to make procedural image files, because that is friggin' beautiful. <3

If I had to add my one-and-a-half cents, it'd be that PNGs can use several types of compression, each based on the difference between any given pixel and the ones that preceded it... it looks like this program might just be mistaken about which type of compression was used?

[davidn]

It's less beautiful when you're trying to get an image of a tree that doesn't look like a post-LSD artist's impression - but that compression suggestion is possible. I've been able to gather than some images in PDFs are compressed with DCT, some with Flate, and some with a mixture of both... my best hypothesis was that the colour model was at fault, but outright swapping it produces the same tearing, just in different colours.

I have a separate library that does it better, so in theory all I have to do is find out exactly what that library does better and repair the source of PDFBox to match... but the methods they choose to drag the image out of the PDF are so distinct from each other in the first place that it's hard to translate them.

[ethelfleda.livejournal.com]

Duuuuuudeee... treeeeee...

[kibet.livejournal.com]

Shame it is not more regular as it may have just needed some line syncing.

It does look similar to images I see in the SEM for samples that are charging. This won't help you at all as it is less of a software issue and more of the scan rate on a sample that is unable to lose localised charing up with electrons.

[pookatimes.livejournal.com]

I've said it before and I'll say it again... it wasn't much to do with LSD.

[davidn]

It was just an Epileptic Tree.

[kjorteo]

 

[lupineangel.livejournal.com]

That icon is perfect for this beyond words. :P

D.F.

[kjorteo]

Thanks!

[starfishchris.livejournal.com]

How bad do you have to be at Yellow to make your Pikachu break?

[kjorteo]

This is Pikachu's "I'm so disappointed in your ability as a trainer that I'm not even bothering to obeying Euclidean geometry at you anymore, let alone your stupid orders" face.