There was some interest in the 40kb thread on how to make interesting low-color graphics.
I posted a few images I did up in photoshop. It takes a certain kind of background to understand this kind of stuff well.
Having grown up with systems like the C64 and Amiga and early PCs, this stuff is how it HAD to be done in the past.
Low color creativity.
http://www.effectgames.com/demos/canvascycle/
The things that you can do with low-color are very impressive.
More info:
http://www.gdcvault.com/play/1023586/8-Bit-8-Bitish-Graphics
32 color poo in loo
16 color king crimson
16 color 50% diffusion
I really like the aesthetics of this
Somewhat related
>>54883787
BEST girl
>>54882727
このセメンデモンはだれ?
2 Color.
GameBoy Color Palette
a e s t h e t i c
Early VGA 16-color eroge was godlike.
If you like this art, do a google image search for PC98 graphics/games
http://www.pc98.org/main.html
PC98 image repo. lots of hentai games lol
>>54884066
Awww man I have to play some VNs now.
>>54884125
>that snatcher
And which game is this in particular? And is it translated
>>54882727
Do C64 palette.
>>54884125
>there will never be a snatcher sequel
oh?
>>54882727
Thanks for this thread op. It takes me back to a time I was an awkward teen with my first computer, an Intel 286, and I'd bought several floppy disks worth of images from a shareware seller I found in the ads section of a popular computer magazine.
Most of the images were of things like cars and landscapes and stuff. I was just excited to see images on the computer screen.
Some pics though were of swimsuit models, some of which you could see the nipples pointing through. One pic was of a woman completely topless!
You've taken my back to that time and filled my heart with melancholy. I wish i could go back there and live that life again. Have fewer regrets and more confidence in myself.
Anyways, offtopic. But thanks once again.
I'm stuck in the goddamned amiga years.
Well, its more fun than modern shit.
>>54882727
Thank you Floens
Four colors reporting in.
>>54882727
>>54882727
i would destroy that woman
>>54886503
>woman
The original
ITT: People aged 16 discover dithering for the first time
>needing more than a kb
>>54883877
>/g/ in charge of dithering
>>54887066
mpv in charge of dithering
>>54887111
oops, I fucked up the rotation
>>54883877
This looks strangely awesome when slightly downscaled in nonlinear light using an undersized tent filter
>>54887210
even better when slightly offset, removes some of the stray pixels
mixed dithering methods
>>54887290
For a sec I thought that was a dick pointing into lain.
8 bits color, not a single fuck was given...
JPEG saved with 0 quality. 4 colors, no artifacts, recompressable infinitely without loss.
>>54889154
Huh. That's weird. I see a little smudge on the edges in browser, but opening it in Photoshop makes it sharp and lossless again.
>>54888824
lol, me too
>>54888824
our mind is corrupted anon
>>54889154
>>54889175
chroma subsampling
>>54886147
Underrated
>>54886088
your screenshot looks fucked up on my screen too, but OP's image looks alright
is that because of the resolution or maybe because of Clover version?
>>54883877
is greyscale really 2 color?
>>54890494
>is that because of the resolution or maybe because of Clover version?
it's because whoever programmed that program doesn't know shit about how to downscale images properly
>>54890516
No grayscale is when all the color channels are the same values
there are different methods for instance green may be weighed more but either way Red=Green=Blue
Easiest method is taking the average of the color channels
>>54887066
what software did you use?
>>54883787
Asuka fag BTFO
>>54887066
so dark
>>54891220
5 colors (black, white, red, green, blue)
>>54883866
Gimp - Obama hope plugin.
>>54890826
imagemagick, GOAT for algorithmic image processing of any kind
>>54891220
mine is faithful to the original, yours is excessively bright
compare against >>54886610
>>54882727
Yo OP i'm a designer, do you have a collection of photoshop palettes i can use for generating this shite?
I could go search but i cannot be fucked collecting them
>>54888824
same
>>54892640
Nvm found a pack here
http://t.umblr.com/redirect?z=http%3A%2F%2Fwww.mazeon.net%2Fdownloads%2FRetro-Color-Palettes.zip&t=MmRhYWMzOTI1MGU3ODlmYmIzN2I2M2E4Y2ZmYzlmMjdkMjI2Y2MyMyxRRzdITWIyMw%3D%3D
>>54892722
Can you run this picture through it? I'm okay if you make the picture smaller if you need to, too..
I would've done this myself but I don't have photoshop on my computer.
>>54892783
Looks ok with the NES palette
http://s84i.imgup.net/NES3479.png
too large to upload
>>54892783convert https://i.4cdn.org/g/1464976752104.jpg -colors 16 out.png
(crop due to small uplink)
>>54892856
thanks man
>>54892907
Thank you too!
The aesthetic is unreal; it brings me years, years back!
>>54892611
i'm using imagemagick as well, but i'm not terribly familiar with it
what arguments did you use for yours?
also, if you know, how do i make >2 color images with the same level of dither as the 2 color output? by default it seems to really prefer to just clamp to the nearest palette color
pic related, notice how the 4-color version looks 'flatter' than the 2-color version as a result
>>54893059
ahem, pic related
>>54892783
here is a 4 bit
>>54886147
>>54891269
8 colors is the lowest you can use to accurately represent a color image
>>54892969
Have you seen the other versions of that image? (night, dawn)
>>54893849
I don't think I have..
The dawn one is probably awesome.
>>54893059
>what arguments did you use for yours?convert original.jpg -colorspace RGB -channel R -evaluate multiply .2126 -channel G -evaluate multiply .7152 -channel B -evaluate multiply 0.0722 +channel -separate -compose add -flatten -set colorspace RGB -dither FloydSteinberg -remap map.pgm dither.png
map.pgm is my palette: (here a 2x1 image containing only black and white, but feel free to customize)P2
2 1
255
0 255
This is a bit longer than strictly necessary, since I also made sure to calculate the luminance in linear light.
If you wanted, it could be as simple asconvert original.jpg -colorspace RGB -remap map.pgm dither.pngwith only small deviations.
(If you leave out-colorspace RGBthen it will dither in gamma light, which produces an image that is very overly bright)
>>54883877
>>54887066
>>54887118
How do I make these?
>>54893825
I'd be willing to bet you could get by with just 4 (RGBK) if you have a smart enough error diffusion algorithm (one that diffuses total error across all channels)
>>54893980
i found setting -colorspace made a difference just a moment ago
not the channel manipulation stuff though, took me long enough to figure out how ffmpeg works 'internally', imagemagick still eludes me somewhat
to do anything advanced in these types of programs, you need to know what's happening behind the scenes at each argument, unlike basic programs where the argument order might not even matter
>>54887066
>>54892611
yours is too dark though. any gamma correction increases the brightness from yours
>>54893825
at *least* get the brightness right!
>>54894237
>yours is too dark
mine on top, yours on bottom, original in center
>>54883787
It's a lot better inverted
>>54894307
yes?
>>54894298
Your shitty image viewer is downscaling wrong
http://www.4p8.com/eric.brasseur/gamma.html
>>54886088
Thank you Floens 2
>>54894307
>>54894344
yours is too bright, I inserted a brightened version of the original for comparison
>>54894391
picture taken with the closest camera within grabbing distance
>>54893849
Search up "imperial boy"
>>54894345
4chan thumbnailer: wrong
gimp: wrong
icecat: wrong
chromium: wrong
correct downscaling is really cool and all, but nothing common seems to do so
>tornado
god tier taste
>>54888824
Same
>>54894391
downscaled to 100x pixels with ImageMagick
>>54894446
>4chan thumbnailer: wrong
configuration error
>gimp: wrong
configuration error (set it to work in linear colorspaces), but GIMP is retarded at colorspaces in general
>icecat/chromium
both horsecock, what did you expect from a browser?
>>54894298
>>54894344
>>54894446
this is not me, pic related is what I get if I exaggerate gamma also
I know what you're getting at but your dithers are overboard
>>5489126
>Failed to show image, out of memory
Wow, I can even open those 10000x10000 on Clover but not a 1.4mb gif.
>>54894482
>downscaled to 100x pixels with ImageMagick
notice how 4chan visibly fucks up the gamma in the thumbnail even on this small image.
4chan thumbnails are really not a good way to compare images. view them at full size (100% scaling, no up or downscaling) if you want to be sure
>>54894522
Wtf
>>54894482
>what did you expect from a browser?
better, really, i mean, normalfags probably view more imagery in their browser than any other program
and i mean, c'mon, google even bought the vpx codec, surely they can figure out image scaling in their own browser
what's the reason for it? do you know why they all get it wrong?
i only have a vague understanding of gamma in digital images
>>54894584
>better, really
good luck with that. I mean browsers still turn this >>54883877 into this >>54887247 when downscaling
(pic related is imagemagick for comparison)
>what's the reason for it?
this particular problem is caused by downscaling in nonlinear (companded) light rather than linear light.
To downscale in linear light requires two extra conversions (sRGB -> RGB before downscaling, and RGB -> sRGB after downscaling), so a naive programmer who is not aware of this simple trick will fail to implement it. In general, whatever browsers do is whatever's easiest to implement (and requires the least knowledge).
Note that this only affects downscaling, not upscaling. For upscaling, you actually actively want to be using nonlinear light - since the linear light conversion amplifies the characteristic ringing artifacts from upscaling. The reason it's relevant for downscaling is because when downscaling, you are mixing multiple pixel values into a combined result. Whenever you mix colors, doing it in linear vs nonlinear light is a huge fucking difference
Since light in the real world is linear (that's what pretty much linear means - as in the real world), mixing colors in linear light causes them to behave as if you had just mixed together two different light sources in real life. Mix black and white evenly in linear light and you will get a color that is exactly half as bright on your display (physically)
This is why linear light is important for dithering as well. A white-black alternating checkerboard, on your display, produces about half as much light as a fully white image. So in other words, it will match the color that is exactly halfway between black and white in *linear* light, which is very d ifferent from nonlinear/sRGB.
>>54894748
>So in other words, it will match the color that is exactly halfway between black and white in *linear* light, which is very d ifferent from nonlinear/sRGB.
Simple demonstration
Left is a value that is coded as 127 (50% in sRGB) everywhere
Middle is a perfect black/white checkerboard
Right is a value that is coded as linear 50% (coded as 187 in sRGB)
Notice how from a distance and when unscaled, the middle and the right are indistinguishable to the eye. However, in the 4chan thumbnail (and when downscaling with non-linear programs), the brightnesses suddenly change
In a program (like imagemagick, mpv, madVR, etc.) that does downscaling in linear light, downscaling the image will *preserve* the apparent brightness of the image perfectly.
Also notice that brightness is also preserved when upscaling (usually)
>>54894825
>Notice how from a distance and when unscaled, the middle and the right are indistinguishable to the eye.
Oh, and assumed a reasonably well-calibrated monitor. If they deviate somewhat, your monitor is not well-calibrated
>>54894748
right, so basically;
- linear is important for regular unscaled display, since it mirrors reality
- nonlinear is important for storage, as it makes more efficient use of bits
and the problem is that somehow everyone is forgetting to convert between the two during downscaling, despite the potentially large error that results?
how does this even happen
>>54894848
>and the problem is that somehow everyone is forgetting to convert between the two during downscaling, despite the potentially large error that results?
*exactly*
there are actually more ways everybody fucks up scaling. gamma is the least of the problems.
More importantly, everything is using bilinear scaling, and everything is also forgetting that filters need to be *widened* when downscaling (otherwise you get moiré artifacts)
Pic related is what happens when I downscale to 50% (or so) in firefox. Notice how it's half-left and half-right, simply because the unwidened bilinear sometimes decides to be more like the left and sometimes more like the right.
It's fucking terrible
>how does this even happen
Software devs that are not image processing experts are bad at image processing
>>54894845
>tfw when you only see two rectangles and don't get the post
>>54894893
imagemagick for comparison as usual
btw, here's how I'm downscaling theseconvert https://i.4cdn.org/g/1464985374391.png -colorspace RGB -filter mitchell -resize 50% -colorspace sRGB down-im.png
First we transform to RGB (that's imagemagick's name for linear light RGB)
Then we downscale using the orthogonal mitchell filter (which works exceptionally well for downscaling in my experience).
Finally we convert back to sRGB before saving
>>54894893
>Software devs that are not image processing experts are bad at image processing
then they shouldn't be writing image processing software
>>54894934
Meh. Firefox devs still haven't even figured out how to color manage webm
(they figured out how to color manage still images and CSS, but not webm - which is why every time I open a webm in firefox the colors are just blaringly wrong on my monitor...)
Chrome devs haven't even figured out that color management is a thing that exists in 2016 at all...
Which image viewer shows images the most accurately of them all, image guys ITT?
>>54895093
That one built in on Windows 95
>>54895093
I use mpv, which is accurate for images which do not rely on embedded colorimetry or ICC profiles. (And I'm thinking about ways to get support for those into ffmpeg/mpv as well)
For images that *do* contain embedded profiles or colorspace information, I use my own fork of ‘sxiv’ which does accurate color management but has terrible up/downscaling and basically no support for high bit depth images.
in general, you're fucked. nothing's perfect at least in terms of the ones I'm familiar with
I could write a checklist or something though (maybe not now)
>>54895093
... mpv
>>54895190
Well shit, maybe I should start viewing my photos with it.
>>54895235
That's seriously what I do
https://gist.github.com/haasn/7919afd765e308fa91cbe19a64631d0f
>>54886088
Clear some space ya cunt
>>54894748
some more fun with mixing colors in various colorspaces
From top to bottom:
>sRGB (naive blending)
>RGB (linear)
>CIE Lab (quasi-perceptual colorspace)
>CIE LCH (polar variant of Lab)
>CIECAM02 (more advanced perceptual colorspace)
every single image processing program on planet earth will produce gradients of variant ‘sRGB’ on this image, since they just naively interpolate between the two colors. go ahead and try it out
>>54894142
black yellow cyan magenta
>>54895335
also notice the characteristic and distinctive “darkening” of the red/green mix in sRGB. Instead of being yellow, it turns to some sort of ugly sludge
(this is exactly the same effect that makes downscaling in sRGB light produce darker images)
>>54895339
CMYK can't be a perfect reproducor because with CMYK in an additive process you can't dither a fully red image no matter how hard you try
RGBK can do it. you only need a good enough dithering algo
>>54895190
Post the original
>>54895390
it in >>54894345
>>54895404
>outdated Chromium version (30)
>works excellently
Jesus.
>>54895466
you didn't downscale it, dumbass
zoom the page out (ctrl+scroll down)
>>54895404
Isn't that bad
>>54895480
welp
It looks just like in the thumbnail, overlapped.
>>54895531
then it "sucks"
>>54895404
>>54895190
here's another good one to try
>>54895585
here's a reference (imagemagick, linear mitchell)
for best chance of “fun”, use a highly non-integer scaling ratio (I used 456x456 here)
>>54895611
It's actually even more fun if you go down to really small sizes
Just made a low color image and then edited it after, I really like the result
>>54895625
here's a “bad” example
(-filter point)
>>54895640
This is the original image
>>54895645
this is what I expect most browsers to look like
(constructed viampv rings_lg_orig.png --geometry 123 -vo opengl-hq:no-icc-profile-auto:no-deband:no-dither:dscale=triangle:no-correct-downscaling:no-sigmoid-upscaling:no-linear-scaling)
amazing just how much I have to *disable* in mpv to reproduce this result, isn't it?
>>54895694
This is the same but with the processing done in linear light
>>54895705
this is the same but with the filter widened
>>54895721
this is the same but with the kernel replaced by mitchell
>>54895736
and finally, this is the result with 8-bit dithering
also as you can see, it's filter widening (or lack thereof) that are causing the bad results in >>54895190
>>54895751
meant to link to >>54895585
>>54895190 is caused by linear vs nonlinear (as discussed)
>>54895363
how could black red green blue do light colors?
it needs white for 5 total
>>54895846
r+g+b = w
see: your monitor
it's just that your ditherer doesn't seem to understand this
>>54895891
show me a correct 4 color
>>54895846
This is how you simulate white using only RGB
>>54895927
while also doubling as a fantastic way to demonstrate just how shitty image viewers are at viewing images
firefox turns this into a yellowish matrix of blurry dots when downscaling
>>54895927
black, red, green, blue is not an adequate representation of color space, white provides that.
I'm waiting for >>54895917
>>54895927
looking at this image I suddenly realize what the limitation of 4color is: your dynamic range goes down, because the amount of light you can output is limited
That's what's causing all of the stupid effects when you naively try applying the dithering algorithm to it. The problem is not the dithering algorithm, it's the dynamic range clipping!
Pic related isconvert original.jpg -colorspace RGB -fx 'u/3' -remap map.ppm -colorspace sRGB 4color.png
>>54896080
compare against the original image with reduced contrast
it's a bang on fit. The only problem is that you need a brighter monitor to get the same result (3x as bright in fact)
>>54896080
>>54896106
and the great thing is, I can demonstrate my theory too
I increased my right monitor's brightness to compensate and took a photo. (The camera quality is not that great, it looks even closer in reality, via eye. The brightness change also totally threw off my calibration)
either way, original on the left, my 4color on the right
>>54896225
Oh, the gamma also distorts itself quite badly because my display is based on LCD technology and therefore has a quasi-constant contrast ratio, whereas my algorithm is tuned for the same black point - mine actually goes up
that's what's causing the sleeves to appear slightly brighter than on the left, it's simply the effect of completely destroying my calibration and gamma by overbrightening my display
(those people who leave their displays on stock brightness are terrible...)
>>54896225
all you are doing is compensating for the fact that your 4 color palette only represents the dark half >>54895977
>>54896225
here's what i get using;convert 4color.png -colorspace rgb -resize 500 -fx 'u*3' -colorspace srgb out.png
Test
>>54896283
you should try using a magnifying glass to view your monitor as it displays white sometime
you'll notice it doesn't directly display white at all
Can someone explain how/why this happens? >>>/int/60317256
>>54896481
Plot twist: he's using a WRGB OLED panel.
>>54895690
>>54895640
Age of ultron
>>54889175
Firefox doesn't show images in 1:1 pixel ratio anymore
>>54893059
convert -list dither
convert -dither x -colors y kek.gif
http://www.imagemagick.org/Usage/quantize/
convert how-to-draw-pop-tart-cat-nyan-cat_1_000000008932_5.jpg -dither FloydSteinberg -colors 8 8-color\ nyan\ cat.gif
>>54897560
>>54891320
Rodneyyyyyy!
Gayfaggots ITT cant emulate old systems accurately
>>54898822
What's the point of all this?
>>54898951
To avoid focusing on your crippling depression
>>54896780
Since when?
And this is why auto updates are bad.
Some autistic kid joins a team, pushes an update, and ruins countless things for forced updaters then tries to argue their change is somehow superior to the old method.
Then they look up statistics seeing that 90% of users use it, therefore must "prefer it" and say you're a minority for noticing their worse shit.
>>54895190
It's funny because mpv seems to have the largest frame in there, yet you claim to scale the image down
Anyway, how do I do this color pallete shit
this thread is a lot of bs
like always 4chan never ceases to amaze me
>>54882727
>>54882737
>not barefoot
This kills the boner.
>>54895190
I can't believe I play all my media with mpv. Seriously.
>>54899929
i scaled down mpv just slightly, it doesn't really matter by how much, you can see the "sucks" text is already partly blending into the background, which it's supposed to
>>54895550
here's a really small downscale
>>54886610
Sauce please? Who is this semen demon?
>>54896780
Firefox will show that image with fuzzy edges even at 1:1 scaling. It's called chroma subsampling
The author of that JPEG should have used 1x1 subsampling when compressing.
>>54899117
>Since when?
I assume he's probably talking about high-DPI support, which you can forcibly disable permanently by setting layout.css.devPixelsPerPx to 1.0
>>54892687
>>54894460
all this same fag
If you guys are interested in shit like this then you should check out dukope's new project "Return of the Obra Dinn". Whether you play the game or not the devlog is pretty neat, as he talks about all of his dithering algorithms in pretty good detail. Not shilling btw
>>54904489
it's doesn't matter if it's exactly 50% or not
>>54902796
>>54904551
but wouldn't this mean the scaling software actually sucks since it isn't representing the picture as it is?
I am hella confused, than again majority of the images here are mostly "tricks" being played on your eyes kinda.
>>54906426
>but wouldn't this mean the scaling software actually sucks since it isn't representing the picture as it is?
the original image as-is shows only the “RULES”
>>54906462
no it doesn't it has an embossed sucks behind it?
so isn't wrong for it to magically disappear?
btw how is photoshop scaling software, bad or good have you tried it?
>>54898875
>>54898850
>>54898836
>>54898822
>>54898806
>>54898710
>no NES
Fucking disgusting.
>>54906578
Adobe Photoshop CC 14.0 (2013): all scaling modes "suck"
>low
>color
>creativity
>40KB
>>54886863
4chan must make reaction pics lower than a kilobyte.
1/6
>>54906753
I don't get it, why is it that even photoshop doesn't care about this?
as much as I want to believe you are right and shit, there is something off about these theory with the image scaler being bad.
this thread is a bit too autistic
4/6
5/6
>>54898710
>>54898806
>>54898822
>>54898836
>>54898850
>none of them in the original resolution
>no zx spectrum
>no atari 8-bit
>no amstrad
>>54906995
>no NES
>no SNES
>no PS1
>no DOS ASCII
>>54907049
>>no PS1
What's special about that?
>>54907049
>no CGA, EGA, VGA
>>no SNES
>>no PS1
Well, their color pallletes aren't very limited.
>>54906578
the "sucks" pattern is designed such that it *should* blend into the same color as the background
where incorrect scaling causes the "rules" text to blend into the background color
if you look at the unscaled image from a distance, or squint your eyes, you'll notice "rules" stands out the most
>>54907074
>>54907081
Well my bad,didn't know PS1 and SNES are complex. NES isn't anyways.
slowly zoom in and out
>>54883896
Tfw I grew up with that kind of graphics and fapped to it
>>54882727
Don't thank me.
https://imgur.com/a/oNcxG
https://imgur.com/a/oNcxG
https://imgur.com/a/oNcxG
>>54906578
>no it doesn't it has an embossed sucks behind it?
Depends on what you mean by “embossed”. If by “embossed” you mean “readily visible”, no.
The “SUCKS” is designed specifically to disappear into the background when viewing it unmagnified. In fact, on an ideal monitor (such as high-DPI LCD or CRT), you wouldn't even be able to tell it was there.
>>54891269
Wow, this image seems to be exposing every image rendering flaw in chrome.
>>54883866
Gaaaaaaay
OK. But where are the feet?
>>54900344
>>54908821
>footfag
>2016
>>54883787
This triggers my FRC.
>>54907378
>The “SUCKS” is designed specifically to disappear into the background when viewing it unmagnified. In fact, on an ideal monitor (such as high-DPI LCD or CRT), you wouldn't even be able to tell it was there.
Which is what I find wrong, shouldn't you be able to see pass the tricks on these pictures with a good scaler?
anyway why hasn't anybody used better scalers than?
>>54882727
>>54907371
i want to do nasty things to her :3 she's adorable, the green haired one
>>54909301
>Which is what I find wrong, shouldn't you be able to see pass the tricks on these pictures with a good scaler?
A good scaler should preserve the psychovisual effect and appearance of an image. (After all, that's why all good image processing algorithms try to model the way the eye works more than anything)
If an image contains a dithering pattern that looks nearly indistinguishable to our eyes (at a reasonable distance), then the scaler should preserve that property.
>anyway why hasn't anybody used better scalers than?
>>54894893
>Software devs that are not image processing experts are bad at image processing
Also, for more proof of why 99% of programs are wrong and mpv is “right”, see e.g. >>54895585. In most programs, downscaling that will produce ridiculously bad results that in no way represent the original picture at all, even though it's a “fairly harmless” image that does not rely on dithering or ideal monitors to work correctly (and indeed, high frequencies very regularly occur in natural photographs).
>>54906886
this
this thread is pure unadulterated autism
>>54909301
if things didn't blend together visually in real life, it'd be harder to make things such computer displays, which depend on this effect to create the appearance of more colors
for example, a typical LCD only has red, green, and blue elements, and yet in normal usage it appears to produce more than just shades of those three colors individually
there's also an information theory standpoint, the pattern of 1 pixel lines in that rules/sucks image is already at it's limit, that is, you can't display those lines individually anymore as soon as you attempt to make the image any smaller, so they must be blended together or dropped entirely
dropping them entirely is what a point scaler does, and in general that looks shit, but it's fast, so some older/weaker systems do that
blending is a more natural method, which involves mixing the pixel colors together to create fewer pixels whose colors are a product of the original pixels colors
>>54909755
>blending is a more natural method, which involves mixing the pixel colors together to create fewer pixels whose colors are a product of the original pixels colors
Yes, and furthermore blending needs to happen in linear light for reasons discussed in >>54894748 >>54895335
>>54909781
yes, that's specific to how computer images are stored, i'm talking just the reality side
>>54909574
alright than question 2
why haven't they notice this why haven't they updated their scaling methods
maybe its not that important but you guys are making it sound like it is
>>54909860
>why haven't they notice this why haven't they updated their scaling methods
Normies will happily watch video with the wrong aspect ratio (stretched). Do you really think they care about moiré artifacts or color distortions?
>>54909860
>>54909875
Some normies do notice, e.g. https://youtu.be/LKnqECcg6Gw
>>54910018
Re-search a bit on this, but doing blurs with gamma 1.0 is actually ugly looking compare to gamma 2.2
However the smoother blend is better, no wonder my pictures look somehow darker when I blurred them before.
>>54910736
>but doing blurs with gamma 1.0 is actually ugly looking compare to gamma 2.2convert \( logo: -blur 0x4 \) \( logo: -colorspace RGB -blur 0x4 -colorspace sRGB \) logo: -append out.png
Do you really think it looks that much worse? I think the linear light version more closely matches the original's luminance, and is thus more faithfully blurred.
(The nonlinear blur is significantly darker / more intense)
>>54910910
and the same on an inverted (bright-on-dark) image. Notice how the nonlinear version now gets brighter. Linear light version is still the same overall brightness
>>54910956
>Notice how the nonlinear version now gets brighter.
Oh, no, it also gets darker. Of course, since the effect is always the same (colors get darked when blended).
>>54910956
the outer edges *should* be brighter, as you're blending them with the white background
>>54911021
You mean in >>54910910?
They *do* get somewhat brighter, but only when blending in linear light. In nonlinear light they get darker.
>>54910910
>>54910956
The linear results match what you would get if you stuck a blur filter in front of your camera when taking a photo. An actual, physical blur would behave that way.
In fact, I can prove this to myself by taking off my glasses and looking at the (now even more heavily blurred) image. The bottom two look pretty much indistinguishable (in both images), the top always looks noticeably darker.
If somebody has a non-shit camera you could take an out-of-focus image to demonstrate the effect (make sure the image is displayed at 1:1 scaling).
>>54911073
I couldn't set my shitty phone camera to stick to manual focus (taking a picture always made it auto-focus again before taking it) but I *could* make a screengrab while the camera was manually out-of-focus
proof that the middle blur actually corresponds to a realistic, real-life blur operation
>>54911194
unfortunately, there's no way for you to tell if the preview itself is correctly downscaled relative to the image sensor
here's my take (used a cellphone camera in a dark room also, took care of focus by pulling the phone back right after it focussed, then quickly taking a shot)
downscaled with imagemagick
>>54911322
oh, should mention the photo is of the bottom/original image
seemed appropriate
>>54910956
>>54910910
This is from photoshop.
Gamma 1.0
http://i.imgur.com/wxby6XN.png
Default (2.22)
http://i.imgur.com/NeJcifE.png
>>54912160
Just in case you wanna do your own
Gaussian Blur was at 9.0
>>54912160
Oh!
What you're talking about are simply quantization artifacts due to the fact that Photoshop represents images in memory with 8 bits precision.
In other words, you're getting banding artifacts. Try changing the precision to a higher depth (ideally floating point) and repeat your experiment.
(As seen here in krita)
>>54913303
even imagemagick defaults to 16bpc precision
photoshop, are you even trying?
>>54913303
and here with 8-bit linear.
This effect is precisely why you should never clip or store linear values with only 8 bit precision, always use at *least* 16 bit.
ImageMagick has a compile-time configurable value, I think the default is 32-bit (convert --version will print a string likeImageMagick 6.9.4-6 Q32which here means 32-bit precision).
Krita is configurable as can be seen in the screenshot. mpv is also configurable (supports essentially the same options as krita).
If your program deals with color spaces or image processing and does *not* allow you to configure this (or at least use something other than 8 bit), it's a pile of shit
ps. ignore the desaturation, I use a wide gamut monitor and krita was applying color correction, but 4chan strips embedded color profiles from my screenshots so it looks wrong on your end
>>54913340
>Adobe
>trying
we're talking about the company that produces more remote code execution exploits per month than linux has in its entire history
do you really think they're trying when it comes to software?
>>54912176
>that hip
is the girl on the right a snake?
