Image Manipulation Thread

File: china.png (1MB, 1000x791px) Image search: [Google]

1MB, 1000x791px

>>1470346

File: china.png (1MB, 1000x791px) Image search: [Google]

1MB, 1000x791px

>>1470379

File: gy9zyj00.png (2MB, 1565x1750px) Image search: [Google]

2MB, 1565x1750px

https://chan.sankakucomplex.com/post/show/4787816

File: 1500604415596.png (701KB, 850x847px) Image search: [Google]

701KB, 850x847px

>>1470327
You should always optimize your images before uploading them.

File: 1500606116260.png (1MB, 1000x791px) Image search: [Google]

1MB, 1000x791px

>>1470379

File: 1500614200877.png (1MB, 1565x1750px) Image search: [Google]

1MB, 1565x1750px

>>1470599

File: 3dd02b9d45.png (985KB, 1241x1390px) Image search: [Google]

985KB, 1241x1390px

very tru

>>1470629
>1241x1390
Don't. Optimization should be lossless.

File: 1500587133320.jpg (761KB, 1432x1432px) Image search: [Google]

761KB, 1432x1432px

I've been wondering if it's better to increase the portion of hair in the image to emphasize the contrast.

>>1471846
What do you mean?

neat script dude

File: output.png (340KB, 472x357px) Image search: [Google]

340KB, 472x357px

>>1472194
test

File: 1460577022710.jpg (274KB, 1280x720px) Image search: [Google]

274KB, 1280x720px

>>1470327
Can you add one of the shipgirls in the middle?
I'd do it myself, but don't know how you made that

>>1472299
Would be cool to add a "preview" functionality that could display the image in both its states like browsers do. Since it doesn't work the same way in my image viewer.

Also would be cool to add a secondary mode that could be toggled between to get this style >>1472299
using "pngcrush" as a dependency or just falling back to the other method >>1472224 if it's not installed.

I'll probably try to make these changes at some point if you don't. Thanks for the script

>>1472321
>Also would be cool to add a secondary mode that could be toggled between to get this style >>1472299

Well, the difference between the styles is that this style
>>1470353
>>1470780
>>1472299
encodes the hidden image in the value of the light pixels and fixes their density at 1/2, whereas this style
>>1472194
>>1470988
>>1472224
fixes the light pixel values at 255 and varies their density. The former method restricts you to dark colors since when the image is opened, half the pixels are always black. And it does a similar thing to the thumbnail. The latter method lets you use the full range with the restriction that the thumbnail must be uniformly lighter (or darker if you switch the gamma around), but the large contrast between the colors being dithered together (always 0 and 255) is undesirable. There's probably an optimum somewhere that you get by varying both the light pixels' value and density, but I'll have to think and experiment to figure it out.

>using "pngcrush" as a dependency

It isn't needed, the script is already setting the gAMA chunk value.

>>1472427
But you can't use that script to get this style >>1472299
Right? It only does the other one

>>1472450
Right, you'd have to edit the calculation details.

>>1472452
That's what I meant in that other post.

I didn't know you could do it without pngcrush though. I thought I tried it a long time ago with imagemagick and couldn't get it to work for some reason.

>>1472469
With imagemagick you can do
mogrify +gamma 0.023 file.png

>>1471157
I keep on getting that one "no zip" error from PIL when I try to run it

>>1472838
Can you post the full error message?

File: shvnao.png (638KB, 1024x1024px) Image search: [Google]

638KB, 1024x1024px

Next step could be specifying an ICC profile. That gives us the power to specify an arbitrary curve rather than just a gamma value.

File: hssvmx.png (638KB, 1024x1024px) Image search: [Google]

638KB, 1024x1024px

>>1472909
Non-monotonic curves seem to work in Firefox but get ignored by Chrome.

File: ss+(2017-07-22+at+05.34.11).jpg (69KB, 1141x738px) Image search: [Google]

69KB, 1141x738px

>>1472910
Freaky...
Why is that?

File: bap.png (436KB, 700x995px) Image search: [Google]

436KB, 700x995px

>>1472908
somehow it worked

>>1473080
Well, these curves aren't supposed to go up and down like that, only up, so I assume Chrome does a sanity check at some point and rejects the profile entirely.

If you're wondering why what looks like a smooth gray ramp in the thumbnail turns out funny when you open it, what you're seeing are the intensities I told it the color values map to using the embedded profile.

File: funnyhitlerimage.png (672KB, 1000x1000px) Image search: [Google]

672KB, 1000x1000px

Newish heights in shitposting!

File: spoiled.png (61KB, 320x320px) Image search: [Google]

61KB, 320x320px

Also could be used to spoiler on boards without spoilers enabled.

>>1470327

Could this be patched somehow?

>>1475861
Sure, if you used a thumbnail maker that processed the gAMA chunk. Would be a waste of time for most images, though.

Since >>1472224 seemed to be better for the most part than >>1474120 except for the speckles in the hair and the filesize, I've updated the script to use a hybrid algorithm. Also it does a better job of selecting parameters automatically now.

https://gist.github.com/ccd0/519b12828f0fcb4d8f4079387dc4b183

Do mods only get to see the thumbs or soemthing?

File: 1490227363433.png (40KB, 346x323px) Image search: [Google]

40KB, 346x323px

>>1476851
mods don't typically moderate /qa/, newfriend

>>1477139
They do when porn is involved.

File: output.png (160KB, 250x250px) Image search: [Google]

160KB, 250x250px

Testing 2 totally different images

File: 1296175876713.jpg (2MB, 933x3943px) Image search: [Google]

2MB, 933x3943px

Don't know if this or an equivalent has been posted yet, but I want to post it now.

I guess I'll give this a shot.

Bump for bane

>>1476307
If you need accurate sRGB -> linear and linear -> sRGB with arbitrary gamma

https://pastebin.com/mGX7ZDVt

File: akaza (change!).png (3MB, 1024x2271px) Image search: [Google]

3MB, 1024x2271px

big yet unfamiliar
bigger than the biggest bar
think big, that's only half as large
bigger, better, twice as hard
outsized, extra large
velocity: very fast
bigger than the closest star
bigger than the motorcar

>>1478183
Thanks, it should help a little, although I expect most people's monitors aren't calibrated well enough for it to make a difference for them. Especially the LCD users.

>>1474313

>>1478183
Wait, why "with arbitrary gamma"? Isn't the exponent fixed at 2.4? Does anyone use variations of sRGB with the exponent changed?

>>1478183
I ended up just using the formula on Wikipedia. I don't think PNG with a gAMA chunk means "do the sRGB formula with a different exponent"; I think it just means to use a straight exponent. But correct me if I'm wrong.

It doesn't affect much, though. I'm working as if the input images and thumbnails are sRGB (although I think the thumbnails may be slightly different; Wikipedia says the JFIF default is BT.601). The only place where there's a conversion from sRGB to linear is converting from the input image to be seen when expanded to the target values of averaged x^(1/gamma) where gamma what's stored in the gamma chunk of the PNG. I don't do any sRGB-to-linear conversions involving the thumbnail because I assume that if any color averaging is done in the downscaling, it's done naively to save CPU time. But it would be good to test how it actually works.

>>1480165
And looking over it, I'm also guilty of doing a downscale before an sRGB-to-linear conversion rather than after (this is in the part where I'm testing brightness adjustments on downscaled versions of the images to see how far the images will have to be adjusted to make one lighter than the other). I should look into whether doing it properly helps.

File: checkerboard.png (6KB, 1000x1000px) Image search: [Google]

6KB, 1000x1000px

Posting a B&W checkerboard pattern to see how 4chan's thumbnailer handles it.

File: 808080.png (4KB, 1000x1000px) Image search: [Google]

4KB, 1000x1000px

>>1480184
Compared to solid #808080.

>>1480184
>>1480186
And the thumbnails are byte-for-byte identical.

>>1480184
>to see how 4chan's thumbnailer handles it
You can test locally, I'm 90% sure 4chan uses the python pillow library to generate its thumbnails. Just load and resize and the resized image exhibits these behaviors.

>>1480588
>Just load and resize
Sorry I take it back, you use the
>Image.thumbnail(sizex, sizey, Image.ANTIALIAS)
function

>>1480588
>>1480604
The thumbnails contain this string:
CREATOR: gd-jpeg v1.0 (using IJG JPEG v80)
So I guess they use https://libgd.github.io/

>>1480617
Oh, guess I was wrong then. Pillow does seem to generate similar thumbs though.

I've tried all of the "hidden image"s I've saved over the years and they seem to work.

>>1480617
GD would be expected given that 4chan is written in PHP.
Here is the code from the source leak, also using GD:
https://gist.github.com/shanehou/11252285#file-4chan-php-L2292
It seems that at least the quality setting has changed, though, since the code says
>ImageJPEG($im_out, $outpath ,60);
whereas the thumbnails have the string
>CREATOR: gd-jpeg v1.0 (using IJG JPEG v80), quality = 40

>>1480679
Looking at old thumbnails from warosu
https://warosu.org/jp/?task=search2&search_ord=old
they used to contain
>CREATOR: gd-jpeg v1.0 (using IJG JPEG v62), quality = 60
and they had funny sizes, like 251 pixels instead of 250. Also the first one has 252 pixels.

>>1480814
Samefagging some more, hopefully this is of interest to some.

Math checks out for that image:
https://warosu.org/jp/thread/10

php > $width = 250;
php > $height = 250;
php > $size = [424, 358];
php >
php > $key_w = $width / $size[0];
php > $key_h = $height / $size[1];
php > ($key_w < $key_h) ? $keys = $key_w : $keys = $key_h;
php > $out_w = ceil($size[0] * $keys) +1;
php > $out_h = ceil($size[1] * $keys) +1;
php >
php > echo $out_w;
252
php > echo $out_h;
213

It happened because of rounding. It took 250/424 = 0.58962264150943 and then multiplied it by 424 again, obtaining 250 + 2.8421709430404E-14.
Which was rounded up to 251, after which +1 was added for no obvious reason to get 252.

In any case, clearly 4chan no longer uses this wonky size calculating code. Thumbnails now have a max dimension now of 250 (if OP) or 125 (otherwise), not that plus 1 or 2.

>>1480838
<?php
$fname = "input.png";
$outpath = "thumb.jpg";
$out_w = 125;
$out_h = 125;

$size = GetImageSize($fname);
$im_in = ImageCreateFromPNG($fname);
$im_out = ImageCreateTrueColor($out_w, $out_h);
ImageCopyResampled($im_out, $im_in, 0, 0, 0, 0, $out_w, $out_h, $size[0], $size[1]);
ImageJPEG($im_out, $outpath, 40);
ImageDestroy($im_in);
ImageDestroy($im_out);
?>
on my machine reproduces the thumbnail of >>1480184 with only two differences, namely the libjpeg version left in the comment and the pixel density fields (possibly set to a different default due to the libjpeg version difference).

ITT we reverse-engineer Yotsuba

File: diff.png (16KB, 125x125px) Image search: [Google]

16KB, 125x125px

>>1480899
Tried again with >>1477168 and not working as well. Thumbnails differ in the red areas.

File: 1446092729900.png (254KB, 555x399px) Image search: [Google]

254KB, 555x399px

This image has noticeable aliasing in the thumbnail. It would be interesting to figure out if the aliasing in 4chan's thumbnail generator is something we can exploit.

File: pattern.png (2KB, 555x399px) Image search: [Google]

2KB, 555x399px

>>1480959

File: pattern.png (774B, 255x255px) Image search: [Google]

774B, 255x255px

>>1480965

File: same size with checkerboard.png (765B, 255x255px) Image search: [Google]

765B, 255x255px

>>1480968

File: horizontal stripes.png (557B, 255x255px) Image search: [Google]

557B, 255x255px

>>1480971

File: vertical stripes.png (685B, 255x255px) Image search: [Google]

685B, 255x255px

>>1480976

File: thumb.jpg (1KB, 125x125px) Image search: [Google]

1KB, 125x125px

With
>>1480899

>>1480968
>>1480971
give near-identical results while
>>1480976
>>1480978
give slightly different but very similar-looking results.

File: thumb.png (673B, 125x125px) Image search: [Google]

673B, 125x125px

>>1480993
The aliasing pattern is the same (up to expected JPEG lossiness) when the thumbnail is saved as a PNG instead.

>>1481005
relevant code
https://github.com/libgd/libgd/blob/7e3587af06c7f89cb749442a689638241b486afb/src/gd.c#L3427

File: 150029399348.png (3MB, 1397x4876px) Image search: [Google]

3MB, 1397x4876px

death to ylyl

>>1481113
A little too small to even guess what the thumbnail is.

>>1470327

>>1481153
It's a 4chan thread

>>>/pol/135161858
well that didn't work, anyone else have some better ideas?

File: internal screaming.png (163KB, 357x264px) Image search: [Google]

163KB, 357x264px

>>1476318
>>1476320
>>1476327
>>1478313
>all this porn hidden with altered images

You do know what will happen once /pol/, /tv/ and /b/ learn what you've done right? Keeping this up is just asking fro trouble.

>>1481871
Go back there.

>>1481877
Anyone who didn't know about this already has the memory of a goldfish anyway. People are just improving an existing technique.

>>1481884
>Anyone who didn't know about this already has the memory of a goldfish anyway
Or maybe, perhaps, possibly, they're newfags.

can some one tell me this magic in autist

>>1482049
Computers store images as numbers telling how bright each pixel should be. But the numbers don't tell the brightness directly. Instead the brightness is a mathematical function of the number. A common choice is the number raised to the power of 2.2. The exponent, in this case 2.2, is called "gamma." PNG files can contain metadata that tells programs what value of gamma to use to display the image properly. But the 4chan thumbnail maker ignores this metadata. So if you make a PNG file with an unusual gamma value, the thumbnail will be different from the image.

>>1481877
>some of the porn is deleted
>yet they messed >>1472224 >>1476323 >>1478313

Well, I guess this proves that repording porn hidden with image altering is pointless and the mods have to see someone point it out.

>>1482632
I don't understand, please explain more clearly

>>1483787
how much of a fagtron do you have to be to report something like this?

>>1483787
snitches are losers

>>1483793
>>1483794
>buttman weebs

>>1483796
>confessing to ban evasion

Just when I thought this place couldn't get worse.

>>1483799
>>buttman weebs
Oh so it's you. What a fucking faggot.

File: 1499732125823.png (271KB, 305x377px) Image search: [Google]

271KB, 305x377px

>>1483801
So, that's the only thing you can do to make yourself feel better?

Wow, I knew you guys has some sad lives, but come on. I'll always take joy in knowing I don't have to pretend to be an underage girl to function like you guys.

>>1483803
I'll always take joy I don't have to be a cockmongling snitch like you.
I mean holy fucking shit, how many penises did you suck today to be such a fucking faglord? Must be over a dozen.

>>1483808
In the end I'm right and you can't change that.

>>1483810
Just having so many weeb thread with less than 10 posts feeds me. It's proof I'm right.

>>1483796
>>1483810
>panicking about getting called out for ban evasion

>>1483813
I thought you were an ebin troll, ruseman of /qa/ that bfto's weebs but from this post it's clear that you're just a faggot snitch obsessed with /jp/. How dissapointing.

>>1470327

>>1470327

>>1470327

Does anyone want to learn how to do simple image editing so /qa/ can have more content? >>1491251

>>1489492
Do NOT open this if you have a weak stomach.

>>1489492
Please stop.......

>>1489492
so hot

>>1491478
I need a muse before I make art again.
I have no direction to go in so I'm just going to work on making scripts that do things rather than art.

>>1492837
Sachiko is for bapping

>>1495642
That is what's WRONG with this world!

cool tricks

there was another trick posted in a previous thread that allowed you to use png files to hide entire images, which weren't visible in the thumb nail or the opened image at all.
then you could open it in gimp and use anti-erase (alt+eraser) to reveal the hidden image.

forgot how to make them though.

Tiny little tumors

File: 39d42568fc495af57fffc7602ab25481.png (1MB, 1003x1417px) Image search: [Google]

1MB, 1003x1417px

>>1498947
You erase the stuff as normal and then in something like gimp you keep the "color space information" of the erased pixels.
Then you can open it back up and "un-erase" it with erase tool + Alt.
There's a script for this as well, but I need to search for it since I never used it myself.

>>1500926
bit different than the image you posted though, since it was hidden below the entire image, so it wasn't obvious like that one.
it's in the /qa/ archives somewhere, have yet to stumble upon it again, with all the trash clogging up the search feature.

Is visual problem solving done in a similar way as a path-finding algorithm?

>>1500926
>>1501558
http://desuarchive.org/qa/search/text/transparent%20pixels%20color/type/op/order/asc
https://ccd0.github.io/transparency-remover/transparency-remover.user.js

>>1502209
shame it doesn't work in chrome

>>1470327

>>1502299
Sigh

>>1505141
what internet browser do you use?

>>1505279
safari

Going to test something soon. Nothing major involving or scripts or anything, I just want to mess around with images and I want this thread to live until then.

>>1470327

chrome is shit and google is going to attempt to user their upperhand in the browser market to change browser/internet standards for the worse

>>1510269
sad but true

File: mpchc1.png (1MB, 1280x720px) Image search: [Google]

1MB, 1280x720px

test

File: die.jpg (123KB, 1134x811px) Image search: [Google]

123KB, 1134x811px

>>1513590
Oh, hi bump bot.

It's been a while since I caught one of your posts before you delete it.

File: c79783f6ada17fa42355629fce372b6e.jpg (458KB, 700x770px) Image search: [Google]

458KB, 700x770px

Going to be looking into how to read letters and numbers off of images.
Seems like a hassle. Going to have to give this a bit of mental thought.

>>1514667
I caught you posting in this thread again bump bot.

How are jpg artifacts caused?

>>1514099
There are OCR engines that can do this.
https://github.com/tesseract-ocr/tesseract

>>1515613
Why do that when I can learn myself.

>>1515593
Any signal can be broken down into waves at various frequencies and reconstructed from knowing the amplitudes of those waves. It's called a Fourier transform (what JPEG actually uses is a variant called the discrete cosine transform). JPEG compression is based on decomposing each block of the image into its frequency components and then rounding the amplitudes, using less precision at the higher frequencies for which the amplitude doesn't matter as much to the visual appearance of the end image with all the waves added back together. By throwing away precision, it uses less data to describe the image, but that rounding makes a slight change to the image. It's normally not a big deal, but it can become a problem in certain cases. JPEG decomposes sharp edges into high frequency waves, and the rounding errors create waves away from the edge, called "ringing". That's why JPEG isn't suitable for those kind of images. Also if the compression is set to throw away too much data, or you compress the file multiple times, such as social media re-compressing every time somebody uploads a viral picture to a new site, the rounding errors become large enough to be visible.

File: 1502493213199.png (169KB, 626x689px) Image search: [Google]

169KB, 626x689px

>>1516422
Hi bot-3rs.

Please leave here and never come back.

File: 42a100bc1ff1057c714cf4ccccea350d.png (145KB, 500x500px) Image search: [Google]

145KB, 500x500px

>>1515639
So when you set .jpg quality you're adjusting the Fourier block size?

>>1517773
You're adjusting the rounding precision.

>>1515639
My brain hurts

>>1519245
Would probably work better with pictures.

File: 250px-DCT-8x8.png (32KB, 250x250px) Image search: [Google]

32KB, 250x250px

>>1519245
>>1519356
Here are the waves that the blocks get broken down into, I guess that's a start. If you mix these guys in different proportions you can get anything you want. The JPEG file says how much of each of these to use to make each 8x8 block of the image. The trick is in the part where it rounds the numbers telling how much of each wave pattern to use, and most of the numbers round to zero. Also to restate >>1517911, you're rounding to the nearest X where you can choose the X to use for each of wave patterns. Using larger X values -> less accurate rounding -> less quality.

So basically whoever invented JPGs is a genius.

>>1519379
Well, it's not like they invented the idea of decomposing signals into waves, that had been around for a while.

>>1519364
Huh. I understand it a little bit more, but I can't say I understand it. I don't need to, however.

>>1522428
Don't think I didn't see that one.

Filthy stealth bumper.

>>1470327

>>1525332
Bot-rofl, at last we are face to face once more.

This thread was stealth bumped again.

>>1527878
I see you bot-rofl.

>>1470327

One more stealth bump took place in this thread.

>>1470327

Because canvas

>>1530797

File: ran.png (896KB, 900x1440px) Image search: [Google]

896KB, 900x1440px

Visual object recognition lecture
https://www.youtube.com/watch?v=gvmfbePC2pc

https://www.ted.com/talks/luis_von_ahn_massive_scale_online_collaboration

>>1533214
This is basically saying that Google uses captchas to problem solve things their OCR can't.

Why no webp 4chan?

A number of stealth bumps have happened in this thread during my vacation.

>>1539387
Is that so?

>>1539411
Yes.

>no posts for 10 hours
>page 1
I can tell there has been stealth bumping.

>>1539387
the bot is on vacation too

>>1517773
Transform the matrix so that most values are close to zero, then round them to zero