Colourspaces (JPEG Pt0)- Computerphile
Vložit
- čas přidán 9. 04. 2015
- Harry's: www.harrys.com/ use coupon code "COMPUTERPHILE" for $5 off
What's a colourspace and why do we have different ones? It's horses for courses as Image Analyst Mike Pound explains.
Digital Images: • Digital Images & Compu...
Atari ST: Pushing the Limits: • Pushing the Atari Limi...
Professor Steve Furber on ARM: COMING SOON!
Universe of Triangles: • Triangles and Pixels
/ computerphile
/ computer_phile
This video was filmed and edited by Sean Riley.
Computer Science at the University of Nottingham: bit.ly/nottscomputer
Computerphile is a sister project to Brady Haran's Numberphile. More at www.bradyharan.com
![How are Images Compressed? [46MB ↘↘ 4.07MB] JPEG In Depth](http://i.ytimg.com/vi/Kv1Hiv3ox8I/mqdefault.jpg)
YCbCr was also used in color TV transmission for backwards compatibility with black and white tvs, because if you pick up just the Y component you can get a black and white image.
1:50 There's a mixup with names of color models: RGB is additive and CMY is subtractive.
Yeah, because the terminology refers to the light. The dyes subtract light. He got confused because you're adding pigment to make it darker, and more pigment means more light subtracted
This felt like a giant teaser - when do we learn all the tricks, limitations, abilities, etc. of each color space!?
FYI he flipped the terminology at 1:54. Additive process is RGB. YMCK is subtractive.
Brilliant, thanks. CMYK is considered subtractive colours. RGB is additive. ;-)
1:54 it's not additive, RGB is, CMY is subtractive - you're "subtracting" light from white making it darker.
point6000x You're right. An easy way to remember is to ask: How do you get the colour white? Add all colours of light, subtract all ink from a page.
point6000x Wow, you're so smart! Nevermind the fact that he was actually talking about the ADDITION of dye not subtracting of light...
Bob Chap en.wikipedia.org/wiki/CMYK_color_model read first sentence. If that's not what he meant he should have specified better. But it's just a minor mishap, nothing major - just wanted to point it out, no harm intended.
Bob Chap It's clear what he meant but the way he expressed it is contrary to convention. While in the grand scheme it makes little difference, it is probably better to conform to conventions for the sake of educational content imo.
Jimmy De'Souza Colorspaces that behave like light does, are additive, like paint does, are subtractive. This is what the word means in color spaces. He is adding confusion by using the term inappropriately for its intended use.
Somewhere in the middle he said the ink was additive, Its subtractive.
Light is additive :P
Can you also do a video on gammut and the different sizes of color spaces? sRGB vs. aRGB, why YCbCr and CMYK have smaller gammut than RGB, etc? Also: please explain LAB-space or XYZ-space and perhaps melissa/prophoto-RGB?
***** Yes please, do this!
***** I was also hoping for a little more along those lines, plus Adobe RGB, and increasing bit depths, and how RAW images work with all that.
***** YCbCr doesn't necessarily have a smaller gamut than RGB. It is just a way of encoding RGB. Barring any rounding errors or range-limiting, RGB to YCbCr and back is a lossless transform.
***** RGB by itself is inaccurate, because it simply tells the monitor how much red, green and blue light to mix. What it doesn't tell is the exact color of those three light sources. The tone is also an important factor. That's where sRGB and Adobe RGB come in, they precisely describe the base components and the tone.
CMYK is even more critical. Every ink is completely different, and it also depends on the type and color of the paper, and the color of the light under which you view the paper.
XYZ and Lab are physical color spaces, which means they're precisely defined in absolute terms. Lab is just a non-linear transformation of the XYZ space. When you transform between different calibrated color spaces, you either go through XYZ or Lab as an intermediate space. For example, when you go from sRGB to CMYK, you really go from sRGB to XYZ to Lab to CMYK.
Gamut means a collection of all colors that a device can display, or a camera can capture, or a specific image file can carry. Monitors have colors that cannot be represented in paper, and the other way around. Out of gamut means a color that cannot be represented precisely, so it has to be substituted by something else. We call this clipping. It can cause different color values to look identical, which may ruin tones and gradients.
sRGB is important, because most TVs and monitors are reasonably close to sRGB, or near identical to sRGB when calibrated. Adobe RGB is important, because high gamut monitors are close to it. ProPhoto RGB is so wide that it cannot even be displayed using today's technology, as far as I know. It's useful when you want to make extensive tonal editing. However, for display purposes it has to be narrowed down to something smaller.
MichaelKingsfordGray You get an instrument called a colorimeter, and measure various RGB values that the monitor can display. Then create a software profile, which tells your operating system and/or image editing software what corrections it needs to make. I do not propose this, it is how photographers and designers do it.
CMYK is subtractive, RGB is additive
Some mention of relative gamut would be nice too :)
Cool, looking forward to the compression/down sampling part. Will be interesting to hear what clever way they have come up with.
This subject is really interesting to me. I am really looking forward to see more of these videos
Starting my 1st image processing course, so had to get a gist and motivation before starting the actual course. COMPUTERPHILLE ❤️❤️
This is very interesting. I, and it seems many people in the comments, would love to watch MUCH more on this topic. Almost every point he made I wanted more depth.
Keep up the great work.
At 4:24, is it not from -128 to 127 ? I find it strange, a value from -127 to 128
Hundersoft You're right, that's a mistake in the video. And usually you add 128 to the Cb and Cr values to get a non-negative range.
Super interesting video, looking forward the Jpeg one !
great video the best explanation of ycbcr format I've seen!
wow! thanks for doing it!!! great material! keep it up!
This needs a follow up about TV broadcast signals, to see the YCbCr in action.
mfaizsyahmi.
Well, the next video is going to be about JPEG, which uses the YCbCr color space.
mfaizsyahmi.
Well, I guess, but RGB isn’t really that ubiquitous. What happens is that people are less aware of other color spaces. Everything that is printed is CMYK, but how many people do you think know about it? Not many, I’d think.
Judging by the fact that the vast majority (or so I believe) of photos on the Internet are stored as JPEG, I’d say YCbCr is far more common than RGB.
Although... computer monitors display images as RGB, so even though JPEG images are YCbCr, what you see in your screen is always RGB...
*****
and not to forget about RCT and YCoCg.
both are "similar" to YCbCr, but are integer reversible, allowing for a lossless to/from RGB (at the cost of needing a larger number range).
there are also other YUV spaces, which are like YCbCr (and will generally look "about right" if mixed up with YCbCr), and which mostly differ in that the calculations are tweaked.
one example is tweaking the calculations to make them cheaper to calculate, for example:
Y=(5*R+8*B+3*B)>>4
U=((B-Y)>>1)+128
V=((R-Y)>>1)+128
(where 'x>>y' means essentially 'y/(2^y)', or diving by a power of 2, but is a lot cheaper to calculate than using a '/', nor does it have the implication of producing a fractional output...).
There's also HSV/HSL colour spaces. Like Y'CbCr these colour schemes separate the colour and luminance information. However, instead of have 2 colour channels, you have a hue channel (describing the colour from the colour wheel) , saturation channel (describing the intensity of the colour) and a luminosity /brightness value channel (which is the grayscale brightness) . These colour spaces are used in software like lightroom and Photoshop for manipulating the colour in images
These videos on JPEG are all I wanted for Christmas, and I got it for free! Thanks, Computerphile.
Does anyone have a recommendation for youtube videos to teach H.264 in depth?
it is -128 to 127 because in 1 byte, there's an equal number of negative numbers and positive numbers. negative numbers start at -1 but positive numbers start at 0.
Yes and no. :)
1 byte can represent 256 different values. That is all it can do.
The meaning of those bytes is what we want. We can say it is unsigned integer, then it is 0 to 255. We can say it is signed integer, then it will be -128 to 127 as you say.
But it can be anything else, if we want it to. For example if you build some science experiment and want to log data about it for long period of time, like temperature, which you know will be between -10 and +50 degrees with granularity of 1/4 degree, you can store it in a byte...
00000000 will be -10, 00000001 will be -9.75, 00000010 will be -9.5...
It's only about how you define your format according to what you want to store (and have a means to code and decode the information into that format).
So in this case, I would say it's neither from -128 to 127 nor from -127 to 128. Its from "more yellow" to "more blue".
***** in this particular case it's -128 to 127. I'm talking about the part where he was unsure about which one. -128 to 127 is because we interpret the byte using two's complement. If we use sign-magnitude, we will have -127 to 127 with 2 different zeroes: 0 and -0.
0 is 0, not positive.
If you are encoding a signal as a signed integer, you would typically ignore the extra negative value, so that the positive and negative limits are the same. Otherwise your signal would end up with a DC offset.
A Computerphile video on jpeg (maybe gig, tiff, etc). Can't wait. I think they already did one one how to pronounce GIF.
Nice video. I found it very interesting I'm looking forward to hearing more about .jpg
Great video!!
I remember a time in art college when the lecturer asked us to predict what would happen when he shone a blue spot-light on a red screen. Very counter-intuitive!
This is very coul!
Excellent video
Great videos!
Very informative. Thanks.
i just had this in school. this video would have been very helpful (cool video)
If anyone wonders why it's CMYK and not CMYB: the K in CMYK stands for Key, not blacK. It is because C/M/Y plates had to be aligned/"keyed" to the key/black plate.
Also, generally printers don't always print in pure K for the blacks. You get a much richer black by also adding various degrees of CMY (depending on the shade is needed).
JPEG works just fine without subsampling the chromas. It achieves compression by _decorrelating brightness from color_, and can efficiently store greyscale and saturated parts of the same image. (For example the flower on a grey background, or a page of black/greyscale text with a few photographs.) Chroma also tends to have less information in it naturally, of which JPEG can take advantage of dynamically, block per block (via the frequency domain transform, not subsampling). The picture of the flower looks quite flat and uniform. Although it is also partly because of how the eye perceives them. They are "easier" to look at an analyze if interpreted as greyscale.
Unlike JPEG, the PNG format doesn't do any decorrelation of color. Greyscale or near-greyscale images will be three times the size. I think they would have to do some clever tricks to have lossless compression at YCbCr or YCoCg, because part of the range of the chroma values corresponds to impossible colors like negative light or saturated white. So more bits would be needed for the transform to be reversible, increasing the file size, which partly defeats the purpose.
Color subsampling is JPEG "abused". Loook how ugly the tumbnails to the right look, how every red bleeds into green, and blue detail becomes unreadable. Full resolution JPEG is perceptually lossless for everything except gradients. Subsampled is cheap and good enough to feed to the consumer (as Google would see it).
Correct if I am wrong but it's Luma not Luminosity in YCbCr. Luma is device dependent. Luminosity (or Luminance more correctly) is independent.
Will the video about JPEG also include dithering? If not I think it would be a great topic for another video.
fun fact :
flip the bits and add one
- that's how computers deal with signed number conversation.
0001 or 1
1111 or -1
Its ingenious design cuz try 0 :)
I hadn't thought about that in a long time, ingenious.
1'51" , CMYK is not working in additive way, it is working in subtractive way. Everything subtracts from white.
On another hand, RGB works in additive way, we call it additive mixing.
Is there a colour space for tetrachromats?
How would it work?
02:19 when you use cmy combined to make black it's called process black and is kind of weak and muddy. Add k (black ink) and you get true black.
CMYK is subtractive, not additive, as you are subtracting the from white reflected light. RGB is additive as you are adding light to an unlit surface...
On the model at 5:08, if luminance is zero, shouldn't the resulting colour be black (no matter what the Cb & Cr)? So the bottom plane should be completely black, right?
ive readed that sometimes the Y component gets more than 1Byte (255 range) because the eye is most sensible to it. and the Cb&Cr components gets less.
so you achieve something more detailed to the eye with the same byte weight/pixel.
i guess the fact of different pixel sizes is irrelevant because you can have your own density in any format, but the convenience of the YCbCr (or at least YUV iirc which is similar) is that you "lose less" (see less loss sssss), and he didn't mentioned why you use YUV instead of RGB or CYMK. Also i think that old TVs (catodic rays) physically needs YUV encoding of the data for the way it works (just like CMYK is because of ink).
I don't remember correctly, but i assume the guy knows what he's talking about.
I'm a bit curious about the geometry (linear algebra) of the color space. I get the sense that the base colors will likely not be perpendicular (in the inner product sense), so is it really correct to call this a cube? I feel like this must be some weighted and sheered space, but I might be wrong.
MichaelKingsfordGray Our color spaces are not in any reasonable sense an approximation to our receptor stimulation responses. That we use three primaries, and are also trichromatic, is about as close as the approximation gets.
A receptor based color space would look qualitatively similar to CIE-1931; most of the colors would be missing, and the visual color gamut would form strange shapes (this is due to the overlap of cone sensitivities). If you include such colors, colors would likely not have unique coordinates (because color perception begins with the opponent color process, which works off of comparative signal strength rather than pure signals).
MichaelKingsfordGray I'm not exactly a specialist, just interested in the subject.
"What is the colour space for human tetrachromats?"
That would simply be LMS color space.
MichaelKingsfordGray Ah, my apologies. Tetrachromatic vision is extremely rare--as far as I know there's only one confirmed case. I'm not sure it's been studied enough to create a color space model for it.
What does he mean when he says down sampling ?
The most practical reason you need Black (K) in printer is because of text. If you print fine text with 3 colours, you can expect it to come out as a smudge because you need to overlay it 3 times. Even with high-precision offset printers, a small pixel of mis-alignment or machine error will "blur" the text.
That is also why designers tend not to use CMY-mix colours on fine body text (like newspapers and books).
Great video!!!!! Why not just use Y and then 1 code representing hue on a line?
***** If you're using luminosity and hue, you also need to add saturation (HSL or HSV color space). Saturation makes the difference between muddy, pastel and vibrant/cartoonish colors.
Another big reason for TV to use YCbCr is that black-and-white TVs can work perfectly fine by just taking the Y and ignoring the rest. Not so important now, but backwards compatibility had to be maintained at every step of the way from the first broadcasts to the modern day.
'Downsampling' of the colour components = using a lower sampling rate?
Does that mean the color components have larger pixels than the luminance component?
HeyHeyHarmonicaLuke Yes, lower resolution = larger pixels.
The possible values for 7 bits is -128 -> 127
Can you do a video about probabilistic programming?
What's the difference between YCbCr and La*b* color spaces?
Is he talking about colorspaces (Rec. 709, Rec. 2020, sRGB, Adobe RGB) or color models (RGB, YUV, Y'CbCr, CMYK, HSB, Lab)?
Oh, I like YcBcR. This is new to me. I feel like, though, it's better for machines to understand, not so much people.
Um a printer is Subtractive and a monitor is additive its about how light works not how you mix the colors.
Any one know how I online order this papers?
Another reason for a separate black cartridge is that C ink + M ink + Y ink is more expensive than using K ink.
What editing software is shown at 2:41?
whats the difference between ycbcr and l cie l*a*b* ??
YCbCr is a system I haven't heard of yet. I'm curious, because RGB and CYMK use three colors each because that's how many colors you have to combine, minimum, to get all the visible colors. But Cb and CR each can't be represented by a single color at some intensity. So why have two? Why not just use Y and then one code representing hue on a line?
klutterkicker Because that system wouldn't have any way to represent any kind of gray. No 25% gray, no 50% gray, no reddish-gray, no grayish-blue, nothing. To cover all of those colors, you'd need to add a third dimension. You could call it, say, "saturation", and boom: you've just invented the HSV color space. HSV is a valid and useful color space, it's just different from YCbCr.
klutterkicker Because that's not enough. RGB is a three-dimensional color space. If you want to transform it to an equivalent representation that doesn't lose too much information, it also needs to be 3D. Even CMYK is 3D, despite having 4 variables - the K is not linearly independent and therefore does not add a new dimension. Your example of Y plus hue is lacking the dimension of saturation. You could represent a bright red (max Y, hue at red, implicit max saturation), but not a desaturated pale whiteish red (max Y, hue at red, low saturation).
And by the way, those color spaces usually can NOT represent all colors we are able to perceive.
if y is blackness, how can we represent green 256, 0, 0, when we decrease cb and cr we would reduce the red and blue component, but only by 127, or is ycbcr have -127, 128 green, which would make the center 0, 0, 0 thus black, if so, why is the y not called cg
So I read some where about Color Dimensions, on April 1st, I thought that was going to talk about that XD
Converting to YCbCr isn't just fundamental to JPEG compression, it's essentially the standard for all types of video coding. So even this video we're watching is likely a VP8/VP9 stream getting decoded to YCbCr and then converted to RGB in or just before the Flash/HTML5 canvas.
So why use YCbCr instead of HSV?
What on earth was that noise he made at 6:28?
That was a rather brief treatment of the subject ...
Penny Lane And that upsets you... I can tell...
Penny Lane Buy a book then, or sign up for a uni-course that goes through this.
Völundr Frey Good lord, what have I done to you that you treat me like that? You never know who people are on the internet, do you? So the benefit of the doubt is more prudent than anywhere. I happen to have a PhD in computer science. So how do you think you basically telling me to go read a book makes you sound like? And based on _what_? The sentence "That was a rather brief treatment of the subject." Seriously.
It's meant to be brief. They're giving a general overview of the subject, not a university lecture.
Penny Lane For future reference: don't criticize anything on youtube, or you will be bitten.
But what of the color *out* of space?
Gareth Dean there non-visible.
***** also they are not well defined. for example how do you use a scale baste on blueness to show a color that is more blue than blue.
*****
And let's not get started on octarine.
MichaelKingsfordGray
That movie..
Oh the pain... the pain...
*****
You'd be surprised. Ultraviolet is reported by those with Aphakia (No lens in the eye) as the blue-sensing cones in the eye are sensitive to these wavelengths but they are excluded from the eye by the lens. Likewise our red receptors are sensitive to some of the IR spectrum but as the wavelength gets longer you need a higher intensity for the light to be visible, eventually reaching the point where damage occurs to the eye. This is a natural result of the way our color receptors work.
is no one going to talk about the three perfect circles drawn equally sized and far apart from each other... by hand! 1:00
So why does this not have anything on HSL colour?
almost no one uses HSV or HSL
How printers achieve orange?
octavio echeverria Y: 100%, M: 50%, C: 0%, K: 0%. Also, it might seem funny at first, but red is 100% Y and M. EDIT: Wait, that's shown in the video anyway, whoops.
Cb and Cr are from -127 to 128? Why not from -128 to 127. 8-bit Signed integer in computer science is from -128 to 127.
CMYK on a printer is the average of colors.
Cyan = (0, 100, 100)
Magenta = (100, 0, 100)
Yellow = (100, 100, 0 )
Average = (66, 66, 66 ) = Light Gray :)
But with the help of black ink and white paper you can fake quite a lot of colors.
What about HSL and HSV?
What about YUV?
Television uses YCbCr because the original TVs used only the Y, for a grayscale (black and white) image. When color TV was invented, they wanted it to be compatible with all of the existing black and white TVs, so they added Cb and Cr to the existing Y signal.
therealEmpyre That's right, except that it wasn't YCbCr, but YIQ (NTSC) or YUV (PAL). They are similar to YCbCr, but not identical.
B/W TV colorspace: Y =)
Jari Komppa Yeah that's also one of the reasons why YCbCr was made, so it could be backward compatible with old B/W TVs :)
And B&W TV worked so well because the eye is more sensitive to brightness/lightness than it is to colour.
Cmyk is subtractive not additive. That's why you end up with black when you mix them all. Rgb is additive.
I thought CMY combination gives brown
WOW
WHat about YPbPr?
Alex Shannon YPbPr and YCbCr are the same colourspace, YPbPr is analog and YCbCr is digitalized (scaled and rounded)
My television has a YPbPr imput
Lol Spider-Man (tobey maguire) explains JPEG colorspaces. Rad.
_"Between 0 and 255 like RGB is"_
Well, Gegl uses 32 bit floating point per component...
:P
I know what he means though.
***** You can specify in OpenGL a color value using any type from a byte, to a int, to a double, and in GLSL, you can use highp floats for color(which is a waste). The graphics card will always output in whatever the monitor uses, which is virtually always 24 bit color - unless you have some kind of weird deep color monitor. It's a nice feature, IMO.
BinaryHistory GEGL... Not GLSL... Not OpenGL...
***** Yes, I know.
So, actually, they're a shaving product supplier, not a shaving service.
If they were a shaving service, they would send someone to shave you.
They don't. They send products.
Hang on.... You've been using Harry's for a while but they only deliver to the US...
Yes they sent me some 😉
lucky :P
Most of the falsehoods are done by 2:52 :)
Actually cmyk is subtractive, not additive.....
CMYK colors are subtractive.. not additive..
What about HSV?
BeastOfTraal And how it connected to angle..
Skisful Hue, Saturation and Value, another system to describe color in a numeric way. Munsell, an american painter described this system first.
BeastOfTraal HSV and its relative HSL are just transforms of RGB. They're defined in terms of RGB in ways that make it easier for us to pick out the kind of RGB colour we're after.
***** HSV was the first system to be used in NTSC broadcasting for describing color for color TV's in America. Thing is, all color TV's use RGB at the Cathode Ray Tube. HSV was the human interface.
Dirk Simons Actually, no. NTSC uses YIQ colour space, which is a rotated version of YUV (which is itself basically an analogue version of YCbCr). It's rotated so that the blue-orange axis can be given more bandwidth than purple-green (which gets a more aggressive low-pass filter), giving more precision for skin tones, while still saving some bandwidth compared to sending both at the same resolution, as is the case with PAL.
I picked that up in the course of doing some research into how analogue TV worked when trying to write a shader to implement the PAL system and reproduce the glitches.
Nooooo, here I thought deep space was more than I could handle!!!!1
Actually, the reason black is K is that it stands for key plate.
It's not blacK?
@@TiagoTiagoT read the comment you're replying to and you will get your answer.
Pretty sure these are colour models, not colour spaces.
4:13 What is that, "nought"? Is that some british word for zero? .__.
Am I the only one that need's subtitles?
Imagine watching this video and being colorblind
Facebook message at 2:40 :P
andrew b Not hearing it.
it's quiet but noticable. confused me for a few seconds :P
dO i LoOk LiKE i KnOw WhAt A jPeG iS?
You mean color?
At 1:54 he made a mistake; he said _additive_ when the correct term for the CMYK system is _subtractive_. RGB is an example of and additive system. Not tryin to be a dick, just pointing it out.
He's explaining color models but he calls them color spaces.
I am not sure if shaving utensils are a potential market for the viewers of this video. I imagine that a lot of bearded folks watch this channel. Their occurrence here is probably higher than on other channels.
First :)