What's the big difference between YIQ, YUV, YCbCr, etc?
Chris Barnett
just "rotated" versions of RGB. That is, they are all RGB multiplied
by some matrix (an affine matrix in case of YCbCr).
I assume the big advantage of these models is that you have the
luminance ("brighness") information in one single component (Y) and
the chrominance (color) information in the other two, while in RGB
these characteristisc are implicitly represented in all three
components.
I can imagine that having them seperated is useful for image
processing, compression (cause this way you can easily keep more
luminance than chrominance info, since the human eye is more sensitive
to luminance differences), or displaying on black & white TV's (just
display the Y channel). This is also true for the HSV (or HSI, or HSL,
or whatever you want to call it) model, which also has one luminance
and two color components, although the HSV model is ofcourse not an
RGB-rotated variant like the rest.
Is this right so far, or did I completely miss the actual purpose of
those models? :)
Now, what I wonder, are there any essential differences between these
color models? I noticed YCbCr uses somewhat difference RGB-weights for
the luminance than YIQ and YUV. But for the rest, what's the big fun
of these different models? And how about CIE xyZ, Lab, or other
models, are they all also based on one intensity + two color channels?
What are generally their advantages in comparison with the rest?
More specifically, I'm working on some image processing software that
mostly depends on getting accurate luminance information (for example
for edge detection), so I thought using the right color model might
make a lot of difference.
Thanks,
Chris.
Marina Diezler
perhaps I can help a bit about XYZ, Lab & Luv.
The XYZ-Colorspace is some kind of "imaginary" (don't know if this is the
right expression) and was set up because of one disadvantage in RGB. In RGB
one sometimes needs negative vectorcomponents for being able to compute all
the colors the human can see. This is "uncomfortable" :o) and so the
XYZ-Colorspace was set up. There we have no negative values and computation
is easier, for that reason many other colorspaces are based on the
XYZ-space, e.g. HSL/I, Lab/uv, etc.
Lab & Luv were set up because the euclidian distance of color-vectors in
RGB, CMYK, etc. does not give satisfying results on the question how similar
two colors are for the viewing person. So with Lab & Luv the human
color-perception has been taken into computing the color values and in
result we have: if two colors have little Lab/Luv-(euclidian-) distance,
then people assume them quite similar. The bigger the distance, the more
different are those colors. So Lab/uv are quite "psychologial" colorspaces.
Hope I could help you.
Greetings, Marina :o)
Comrade
> As far as I understand, color models like YIQ, YUV, YCbCr, etc. are
> just "rotated" versions of RGB. That is, they are all RGB multiplied
> by some matrix (an affine matrix in case of YCbCr).
there are many RGBs, these ar rotations from some specific RGB, and some
"device independent" RGB of the eye's sensitivity, and not a device
--
Comrade
see my friend Average Joe's site
http://www.mysolution.ws
the aristocracy was the problem in 1776
the aristocracy is the problem today
http://www.aclu.org/dissentreport
we must close the door by which aristocracy arises
"Does God want goodness? or the choice of goodness?
Is the man who chooses bad, somehow better,
that the man who has the good forced upon him?"
a quote from the movie, A Clockwork Orange, Kubrick
Enlightenment is man's release from his self-incurred tutelage.
Tutelage is man's inability to make use of his understanding
without direction from another. Self-incurred is this tutelage when
its cause lies not in lack of reason but in lack of resolution
and courage to use it without direction from another. Sapere aude!
"Have courage to use your own reason!" - that is the motto
of enlightenment.
Kant -- What Is Enlightenment? 1784
Ayn Rand just professed laissez faire, another Jewish mysticist
of Zionist status quo social order, and aspiring to such, she
never challenged "the establishment" she was an radical crony,
not a radical individual
Chris Barnett
"Marina Diezler" <min...@uni-koblenz.de> wrote
>[...]
>In RGB one sometimes needs negative vectorcomponents for being
>able to compute all the colors the human can see.
Ok, but then wouldn't it have been easier to just use a biased or
"shifted" version of RGB? I mean, we could add +100 to all three R,G,B
components (assuming they normally go from 0-255) or some number so
that any color can be expressed with positive RGB values?
And besides, some of those other models actually use negative numbers
even to represent "positive-only" RGB colors. YIQ and YUV for example
(if I'm not mistaken).
>Lab & Luv were set up because the euclidian distance of color-vectors
>in RGB, CMYK, etc. does not give satisfying results on the question
>how similar two colors are for the viewing person.
>[...]
>if two colors have little Lab/Luv-(euclidian-) distance, then people
>assume them quite similar.
Isn't this also the case for RGB (or *any* color model even, except
for the Hue component in HSI/HSV)? I mean, usually the closer two
colors are (in euclidian distance), the more they seem to be similar,
no matter which color model you use? Or is the releation between
similarity and distance more equal amongst all three components for
Lab/Luv (cause I can imagine that in YIQ, an Y difference can be more
obvious than the same numerical difference in I or Q)?
Chris Barnett
> there are many RGBs, these ar rotations from some specific RGB, and some
> "device independent" RGB of the eye's sensitivity, and not a device
Ok, I understand that there can be different RGB models. However, the
models I mentioned (YIQ, YUV and the likes) are usually defined as
rotations of RGB, right?
So I wonder, what is really the essential difference between them. For
example, what are the (dis)advantages of YUV over YIQ or YCbCr? Or Lab
vs Luv?
Dave Martindale
>Ok, but then wouldn't it have been easier to just use a biased or
>"shifted" version of RGB? I mean, we could add +100 to all three R,G,B
>components (assuming they normally go from 0-255) or some number so
>that any color can be expressed with positive RGB values?
You could do that. The problem is that you then get a representation
that has to go through a "decoding" step before it can be displayed.
One of the advantages of RGB is that you can just dump the data into a
framebuffer and get something viewable, without any processing at all.
You also get a representation where black is not zero, which is ugly for
some purposes.
Apparently, Kodak Photo-CD provided a way to store negative values.
That way, it can handle colours beyond its regular gamut, yet the
colours that *are* inside the gamut have non-negative values.
>And besides, some of those other models actually use negative numbers
>even to represent "positive-only" RGB colors. YIQ and YUV for example
>(if I'm not mistaken).
Different representations are designed for different purposes. YIQ and
YUV are designed as relatively cheap ways to separate luminance and
colour information. The Y is then displayed on B&W TVs without the
colour components, so it needs to be reasonably good-looking. And the
separation is also used to allow transmitting the colour with about 1/4
to 1/10 the bandwidth of the luminance.
However, these representations are not convenient for specifying colour
since IQ or UV are not independent of brightness. HSV and HSL are
systems designed for specifying colour, while still being cheap to
convert to and from the RGB that hardware works with. Systems like Lab
and Luv are more complex mathematically, but are designed to model
how the human eye sees differences in colour more accurately.
>>Lab & Luv were set up because the euclidian distance of color-vectors
>>in RGB, CMYK, etc. does not give satisfying results on the question
>>how similar two colors are for the viewing person.
>>[...]
>>if two colors have little Lab/Luv-(euclidian-) distance, then people
>>assume them quite similar.
>Isn't this also the case for RGB (or *any* color model even, except
>for the Hue component in HSI/HSV)? I mean, usually the closer two
>colors are (in euclidian distance), the more they seem to be similar,
>no matter which color model you use? Or is the releation between
>similarity and distance more equal amongst all three components for
>Lab/Luv (cause I can imagine that in YIQ, an Y difference can be more
>obvious than the same numerical difference in I or Q)?
It's a question of *how much* difference in colour corresponds to a
certain numerical difference in the coordinates. In some systems, 5
units of Euclidean distance is a very visible colour difference in some
parts of the colour space, while it is not big enough to be visible at
all in other parts of the colour space. In these systems, Euclidean
distance is pretty useless for judging visiblity of differences, or for
finding the "nearest" colour out of a palette.
The point of systems like Luv and Lab is that the coordinate system has
been "warped" to make Euclidean distances in the space correspond
better to visible colour difference. Thus, a 5-unit distance between
two colours is approximately the same amount of visible colour change
everywhere in the colour space. Saying that two colours match within 5
units actually *means* something in these spaces.
Each of these colour systems was designed for a particular purpose. No
one system would satisfy all of these purposes.
Dave
Dave Martindale
>"Comrade" <avera...@mysolution.ws> wrote:
>> there are many RGBs, these ar rotations from some specific RGB, and some
>> "device independent" RGB of the eye's sensitivity, and not a device
>Ok, I understand that there can be different RGB models. However, the
>models I mentioned (YIQ, YUV and the likes) are usually defined as
>rotations of RGB, right?
Depends on what you're talking about. The *specific* YIQ encoding used
in NTSC broadcasting is a linear transform of a specific RGB encoding,
which uses specific RGB primaries and a specific white point, and where
the RGB values are already gamma-corrected in a specific way. So this
particular YIQ encoding is quite well-defined.
If you want, you could talk about other YIQ-like systems, which might be
transformations of *different* RGB systems, or might use linear instead
of gamma-corrected inputs. But nobody actually uses any of these (that
I know of), so you need to carefully specify that you're using an
imaginary YIQ system if you diverge from what's used in NTSC in any
way. Similarly, the YUV used in PAL is well-specified, and you ought to
say something if you're considering a larger class of YUV-like
transformations of some general RGB system.
>So I wonder, what is really the essential difference between them. For
>example, what are the (dis)advantages of YUV over YIQ or YCbCr? Or Lab
>vs Luv?
These systems differ in many ways. What, to you, would be an "essential
difference"? Do you care about ease of implementation in hardware? In
software? Ease of specifying or mixing colours? Or having a space
where the Euclidean distance between colours is related to the
perceptual difference between colours?
To fully cover what you're asking would take hours of writing. Nobody
is likely to do that for you on Usenet when there are textbooks out that
that will answer your questions. If you want a full understanding of
this, find one of those textbooks. If you have a small specific
question ("What's the difference between YIQ and YUV?"), you might get
it answered here.
Dave
Alan Benn
too - very informative with all the explanation you'd want about what all
the various colour spaces are. As to whether this has any relevance to what
you are trying to do though depends on what you mean by "accurate luminance
information". If by accurate you mean according to the definitive standards
as described by Charles, then you'll need to incorporate some form of
calibration in your images or imaging system. ie measure one or more
standards (greys/whites ?) with known reflectance and calibrate your
measured RGB's to the reference values for these standards (L* as measured
by a spectrometer ?).
However , if by accurate you just mean repeatable, consistent etc then I
don't think it matters how you transform the RGB values to intensity to do
your edge detection. An edge that exists in one colour space will still be
there in another - unless you're trying to find edges on (say) red objects
vs blue objects - then the weightings could matter.
I suppose there could also be effects if you're trying to quantify the
relative strength of edges, e.g. a bright white object next to a lightgrey
as compared to a dark grey next to a black. This could influence whether
you want to use a linear or non-linear transformation - or even whether the
image collection device is linear or gamma adjusted.
Alan Benn
"Chris Barnett" <Christia...@yahoo.com> wrote in message
news:6436f134.03071...@posting.google.com...
Comrade
> Ok, I understand that there can be different RGB models. However, the
> models I mentioned (YIQ, YUV and the likes) are usually defined as
> rotations of RGB, right?
well, usually a rotation from CCIR709 RGB, or CIE RGB of eye, but not any
RGB will do, and device oriented mentality, does not serve open system,
and broad application, and device independent systems
> So I wonder, what is really the essential difference between them. For
> example, what are the (dis)advantages of YUV over YIQ or YCbCr? Or Lab
> vs Luv?
the culimination of modern color science, in color specification is CIE
hSL, these are the perceptual correlates by which artisans have always
specified color, hue, saturation, lightness, and, in a device independent
system, such is the optimal choice, for specification, and even order,
since the first ordering of color by Munsell, if you are building a closed
device oriented system, or propreitary closed system approach, then other
means of color specification and order can be used, such as SWOP CMYK, is
the most popular, in any aspect, as with Pantone, or CCIR709 RGB with
display monitors, etc., AND THERE IS NO GENERAL RECCOMENDATION in the
device independent sense, for a device centered model or system
now, the choice of a color space for a device, is defined by the device,
so I will not discuss this
the choice of space for encoding, is a perceptually distributed space, and
preferably the one with less computational error, from your working space
of color specification, hSL, so, hSL is the optimal space, should you want
to maintain the device independency of the file, if not, any perceptual
derivative of your working space will do, simply the one with the least
computational loss from what device space mentality you are trying to
achieve
now, for a device exchange space, or device interchange space, or profile
connection space, the same rules of computational error mininization apply
as with an encoding space, so hSL is the optimal choice here
now, the ICC, www.color.org has chosen CIELAB instead of hSL, and you must
understand you need CIELUV to derive a saturation correlate, CIELAB has
not chromaticity diagram, and no percetual saturation correlate
and such decisions were made by ICC, in perpetuation of previous models,
like Adobe Photoshop, in order to "court" and "appease" certain "players"
wth vested interests in the CIELAB paradigm, at the beginning, but it is
most logical, that they evolve to hSL at this point
--
Comrade
see my friend Average Joe's site
http://www.mysolution.ws
the aristocracy was the problem in 1776
the aristocracy is the problem today
http://www.aclu.org/dissentreport
we must close the door by which aristocracy arises
"Does God want goodness? or the choice of goodness?
Is the man who chooses bad, somehow better,
than the man who has the good forced upon him?"
a quote from the movie, A Clockwork Orange, Kubrick
Enlightenment is man's release from his self-incurred tutelage.
Tutelage is man's inability to make use of his understanding
without direction from another. Self-incurred is this tutelage when
its cause lies not in lack of reason but in lack of resolution
and courage to use it without direction from another. Sapere aude!
"Have courage to use your own reason!" - that is the motto
of enlightenment.
Kant -- What Is Enlightenment? 1784
Ayn Rand just professed laissez faire, another Jewish mysticist
of Zionist status quo social order, and aspiring to such, she
never challenged "the establishment" she was a radical crony,
not a radical individual, the very anti-thesis of individualism
Comrade
> To fully cover what you're asking would take hours of writing. Nobody
> is likely to do that for you on Usenet when there are textbooks out that
> that will answer your questions. If you want a full understanding of
> this, find one of those textbooks. If you have a small specific
> question ("What's the difference between YIQ and YUV?"), you might get
> it answered here.
--
Comrade
see my friend Average Joe's site
http://www.mysolution.ws
the aristocracy was the problem in 1776
the aristocracy is the problem today
http://www.aclu.org/dissentreport
we must close the door by which aristocracy arises
"Does God want goodness? or the choice of goodness?
Is the man who chooses bad, somehow better,
than the man who has the good forced upon him?"
a quote from the movie, A Clockwork Orange, Kubrick
Enlightenment is man's release from his self-incurred tutelage.
Tutelage is man's inability to make use of his understanding
without direction from another. Self-incurred is this tutelage when
its cause lies not in lack of reason but in lack of resolution
and courage to use it without direction from another. Sapere aude!
"Have courage to use your own reason!" - that is the motto
of enlightenment.
Kant -- What Is Enlightenment? 1784
Ayn Rand just professed laissez faire, another Jewish mysticist
of Zionist status quo social order, and aspiring to such, she
Chris Barnett
>You could do that. The problem is that you then get a
>representation that has to go through a "decoding" step
>before it can be displayed. One of the advantages of RGB is
>that you can just dump the data into a framebuffer and get
>something viewable, without any processing at all. You also
>get a representation where black is not zero, which is ugly
>for some purposes.
Ok, although decoding is necessary for *any* model other than RGB (or
whatever the used device's native model is). But I agree with you that
having a non-zero based system is a lot uglier than having to deal
with negative values.
>The point of systems like Luv and Lab is that the coordinate
>system has been "warped" to make Euclidean distances in the space
>correspond better to visible colour difference.
Aah ok, that makes sense. Thanks!
Chris Barnett
>These systems differ in many ways. What, to you, would be an
>"essential difference"? Do you care about ease of implementation
Differences like the ones you mention, yes :) I mean, in my opinion
YIQ and YUV for example are less "different" than YIQ and HSV, since
the particular advantages of YIQ and YUV over other color models are
more or less the same.
>To fully cover what you're asking would take hours of writing.
>Nobody is likely to do that for you on Usenet when there are
>textbooks out that that will answer your questions.
No, of course, I understand. But your answer made things a lot more
clear to me, I'm getting the big picture now and I think I know what
to look for to learn more about the details. Thanks a lot!
Chris Barnett
>These systems differ in many ways. What, to you, would be an
>"essential difference"? Do you care about ease of implementation
Differences like the ones you mention, yes :) I mean, in my opinion
YIQ and YUV for example are less "different" than YIQ and HSV, since
the particular advantages of YIQ and YUV over other color models are
more or less the same.
>To fully cover what you're asking would take hours of writing.
>Nobody is likely to do that for you on Usenet when there are
>textbooks out that that will answer your questions.
No, of course, I understand. But your answer made things a lot more