Effects of saturation on contrast

Discussion in 'Photoshop Tutorials' started by ronviers, Jul 24, 2006.

  1. ronviers

    ronviers Guest

    It appears that increasing saturation decreases contrast. Is that
    true? If so, why would that be?

    Thanks,
    Ron
     
    ronviers, Jul 24, 2006
    #1
    1. Advertisements

  2. The RGB colorspace mixes color and luminosity information in each
    channel. Therefore, changes in color will inevitably change the
    luminosity as well. Increasing saturation beyond a certain level
    results in a loss of luminosity variation and consequently, a loss of
    contrast.
    In L*a*b, color information is independent of luminosity information,
    they reside in different channels. Therefore, color can be manipulated
    without influencing luminosity and vice versa.
    To learn more about this, try the following. Put a Hue,Saturation
    adjustment layer on top of some image in RGB mode. Do not change
    anything at first, but set the layer mode to luminosity first. Now
    open the dialog window again and move the saturation slider to its
    extremes. You will observe a change in the luminosity of the image
    with a considerable decrease of contrast when saturation is brought to
    a maximum. Now do the same thing in LAB (edit>convert to profile>LAB).
    In this colorspace, you won't see a change in luminosity/contrast at
    all, irrespective of how much you change saturation. However,be warned
    that it LAB it is possible to create virtual colors. There is no such
    thing as a bright green with zero luminosity in the real world.
    However, PS tries to optimize the rendition of such colors by
    compromising between luminosity and saturation. The result is quite
    often much better than what can be achieved in RGB.
    The LAB mode has great advantages for image manipulation. You may read
    more about this in Dan Margulis' book on the topic.

    Peter
     
    Peter Wollenberg, Jul 24, 2006
    #2
    1. Advertisements

  3. ronviers

    ronviers Guest

    I am glad you said this because it made me realize that I do not
    understand how luminance is derived from the R,G,B trio in RGB color
    space. I have been thinking of saturation as the radius of the cone,
    hue at the direction (center out), and luminance as the distance from
    the tip. I guess the bigger the numbers the further from the tip, but
    I think that what you are saying is that for each color value in a
    given channel there will be an associated level of luminosity - and
    that as I increase the value of the color the less latitude I will have
    for changes in luminance? Which I do not understand but at least I
    know what to read up on.

    It seems like the implication of this is that, at least in the context
    of my original question, is that luminosity is directly related to
    contrast. Which makes me wonder what is the relationship between
    contrast and detail. Again you have given me direction on what to
    study.

    This was a great exercise.
    I think you are saying it is possible to edit in LAB then use the out
    of gamut conversion or transformation algorithms in PS to achieve more
    desirable results. I remember reading something about these
    algorithms, perception and absolute seems to ring a bell but I will
    have to try to find the article again. I would love to have the book,
    i will put it on my list, but for now it is out of reach.

    One thing I noticed when I was working though your exercise was that if
    I could somehow take the highlights from the 'Red' channel in RGB
    space, and combine that with the midtones and shadows from the
    'Lightness' channel from LAB space I would have the perfect Black and
    White photo. The highlights from the red channel are so delicate and
    lovely while the midtones and shadows fromt he lightness channel are so
    sharp and rich. I doubt if that would always hold true but in the case
    of the picture I was working with it would be great.

    Thank you for all the great informtion,
    Ron
     
    ronviers, Jul 24, 2006
    #3
  4. That is one way to explain the situation. Another way to explain it is
    to say that the editing tools of Photoshop do not behave correctly.

    Naturally, a control that has a label such as Lightness or Saturation or
    Hue etc should affect to that particular property of colors and to
    nothing else, and should do so exactly similarly no matter how the image
    data happens to be coded (no matter in what ever RGB or Lab etc space
    the data happens to be).
    That is not correct, in the Photoshop the Lab space has three channels
    (Lightness, a, and b) but the Lightness is not independent from the a
    and b channels. In order to demonstrate this I created 256 somewhat
    random patches from the base color Lab==50,0,0 by changing only the a
    and b channels (using the Saturation and/or Hue controls) in the Lab
    mode of Photoshop (with the gamut warning tool active for sadRGB so the
    patches are all inside CRT gamut).

    Here is the result as a JPG:
    http://www.aim-dtp.net/aim/temp/lab-luminance.jpg
    The base color is at the top left corner. The patches clearly appear to
    have way different lightness, some patches appear to be very dark, some
    very "bright".

    Here is the original as a Lab mode PSD document in case you want to
    verify the Lab values of the patches yourself:
    http://www.aim-dtp.net/aim/temp/lab-luminance.psd
    As above, no. Affecting to the Lightness does affect to the hue and/or
    saturation ... I could make a demonstration about this also but it is
    somewhat more difficult to evaluate by the unaided eye.
    Actually no. There are only a couple of procedures that provide some
    marginal benefit in the Lab space. The Lab space has major problems with
    out of gamut issue, even a very slight adjustment can throw plenty of
    colors right out from this universe, the Lab space specifies a huge
    amount of "colors" that are out of the gamut of the human vision and out
    of the gamut of the CIE color model that the current color-management is
    based on. It has a discontinuity point in the transfer function at the
    dark end etc etc. Generally it is the very best to totally avoid working
    in the Lab space.

    Timo Autiokari
     
    Timo Autiokari, Jul 24, 2006
    #4
  5. ....
    IMNSHO this is an inherent problem of the RGB space and not of PS. It
    is impossible to increase the saturation and keep the hue _and_ the
    luminosity constant in RGB, if one of the channels approaches its
    extremes. 255R 128G 128B can only be made more red by decreasing the G
    and B values, and since the luminosity is represented by R + G + B
    the increase in saturation inevitably results in a decrease of
    luminosity.
    When I opened this image and looked at the L-channel, I found it was
    evenly grey. Evidently, the lightness _is_ unchanged in your example.
    Which means that it is irrelevant for practical purposes?
    Well, I use PS to work on real life photographs and not on computer
    created swatches and, working in L*a*b I do get consistently better
    results from these than in RGB. Actually, I was able to obtain
    excellent results from shots which I trashed earlier, when I did not
    yet know about the possibilities of L*a*b and tried to improve them in
    RGB or CMYK.
    I think you didn't read the book, which we are talking about. Margulis
    devotes several chapters to the pitfalls of L*a*b and the "impossible
    colors". He makes very clear what the advantages and disatvantages of
    the various color spaces are and actually shows impressive examples of
    how RGB and CMYK may outperform L*a*b in certain situations, how the
    advantages of all spaces can be tied together for even better results,
    and how the apparent limitations of any of them can be turned into an
    advantage with certain retouching problems.
    I have a very pragmatic approach to my work, and as long as my results
    are superior after a L*a*b treatment within a fraction of the time it
    would have taken to get a mediocre result in CMYK or RGB, I care a
    shit about any theoretical reasoning, why this should not have
    happened.

    Peter
     
    Peter Wollenberg, Jul 26, 2006
    #5
  6. ronviers

    ronviers Guest

    I decided to check the contra positive, which should always be true, of
    the original statement that - increasing saturation decreases contrast.
    That is, increasing contrast decreases saturation. I did this by
    looking at the height of the color channels (wouldn't that be
    saturation?) while sliding the gamma slider to the right in
    'Levels'. This turned out to not to be a valid test since the
    saturation (height) only increased where the tonal distribution was
    being squeezed. Where the distribution was being stretched the
    saturation decreased. I could not figure out how to interpret those
    results. Not being one to let my ignorance get in the way I proceeded
    to do the same test in LAB. To my surprise ONLY the L channel was
    affected. The color channels were not changed at all. This seems to
    back up what Peter was saying both in terms of the limitations of RGB
    space and in potential benefits of editing in LAB.

    Brgds,
    Ron
     
    ronviers, Jul 26, 2006
    #6
  7. ronviers

    Tom Nelson Guest

    You can see an illustration of Peter's point in the Color Picker. Start
    with a totally desaturated color and an arbitrary hue (I'll use 0°).
    Increase the saturation slider and watch what happens to the R, G and B
    values. Photoshop saturates color by decreasing the complement. Since
    0° is a red, the red value stays constant and the green and blue values
    decrease as you drag the saturation slider up. If you started with red
    at 255 and saturation at 0, the desaturated version is white
    (255,255,255). By the time you drag the saturation to 100%, you'd have
    255,0,0.

    Tom Nelson
    Tom Nelson Photography
     
    Tom Nelson, Jul 26, 2006
    #7
  8. That is not an inherent problem of the RGB space, the same will happen
    in what ever color space when any of the channels gets clipped.
    The purpose of this image was to show you that even if the numeric L
    value of all the patches is exactly the same the appearance of the
    lightness (for your vision) of the patches is not at all the same.

    Just look at the image, some patches clearly appear to be much more
    lighter than the 50,0,0 patch and some appear to be much more darker.
    So a change to the a and/or b channel(s) does affect to the lightness
    of the color (but not to the numeric L value of the color). So
    something is very wrong, and it is that the Lab channel that is called
    as the L or Lightness channel is not at all a true Lightness channel.

    Timo Autiokari
     
    timo.autiokari, Jul 27, 2006
    #8
  9. The "height" of the histogram of the color channels does not describe
    the saturation of the colors.
    That control does not change the saturation of the colors but makes the
    colors of an image (that has correct colors) to be incorrect. use e.g.
    the Saturation slider in the Hue/Saturation dialog to adjust the
    saturation.
    Correct, gamma change just makes a correct image to be incorrect, it is
    not possible to assess the properties of colors correctly from image
    data that is incorrect.
    So, you adjusted the gamma slider in the Levels dialog in Lab space.
    Now, please open that Lab image again, then go to the Levels dialog.You
    will find out that in the Channels dropdown the default channel (that
    is to be adjusted) is "Lightness". So, when you adjust the Lightness
    channel then of course only the numerical values on the Lightness
    channel will change. No surprise in that. In the RGB mode the default
    channel in that Channels dropdown of the Levels dialog is "RGB" so all
    numerical values on al the three channels are adjusted.
    The numerical values of the color channels remain unchanged of course.
    But look at the actual colors in the image (as shown by your monitor)
    while you adjust the Lightness channel, both the hue and the saturation
    will change but only the Lightness of the colors (as shown by your
    monitor) should be changing.
    There are many RGB spaces, some of them indeed do limit the gamut
    (saturation) some do not. Some RGB spaces are better for editing
    purposes and some are worse, this depends on the transfer function of
    the particular RGB space.

    Lab space does not (in practice) limit the gamut but for editing
    purposes it is a _very_ bad space due to the extremely cumberome
    non-linear coding of the Lab space and due to the fact that nearly all
    of the editing operations of Photoshop are designed for editing linear
    RGB data only. While editing in the Lab space you will very easily get
    large luminance and/or hue and/or saturation errors even if the
    numerical Lab values do not indicate that at all.

    Timo Autiokari
     
    timo.autiokari, Jul 27, 2006
    #9
  10. ronviers

    ronviers Guest

    I agree that I do see changes in saturation as a result of changing the
    gamma of the Lightness channel. But now all this has me wondering -
    how is it possible to edit in anything other than sRGB since that is
    gamut of the monitor? How can we even look at LAB space using a CRT
    display device? In other words, why don't all color spaces look like
    sRGB?

    Thanks,
    Ron
     
    ronviers, Jul 27, 2006
    #10
  11. sadRGB is just an approximation of CRT monitors but the the worst of
    them. It is a very bad color-space for image editing (it has the
    slope-limiting in the dark-end and it is highly non-linear).

    Editing in a color space that has larger gamut (than what the gamut of
    the display is) indeed gives some obvious trouble. Adobes provides some
    tools that help in this, mainly:

    1) Proofing & gamut warning

    2) The "Desaturate Monitor Color by" -inputbox in the
    ColorManagement/MoreOptions (or ColorManagement/Advanced) section. But
    is it difficult to work with.
    Monitor only shows what can.For CRT monitors a very good appriximation
    is the nativePC profile (Primaries=Trinitron, Gamma=2.5,
    Whitepoint=D65). Color that that are out-of-gamut from that are shown
    by Photoshop with one or more channels clipped (so they are shown
    inaccurately).
    Short answer to this could be: Because we have color-management.

    Now, sadRGB is just one color-spesification among others.
    Color-specifications are the mathematical translations between the
    numerical values of the colors and the visual appearance of those
    colors. For example:

    the code RGB== 50,100,150 in the sadRGB -space outputs a certain color
    on the monitor, that very same color is specified as:

    RGB== 68,122,160 in the nativePC -space,
    RGB==70,100,147 in the adobeRGB -space,
    RGB==31,31,72 in the CIE 1931 d65 gamma 1.0 -space.

    RGB working-spaces normally/currently consists three parameters:

    -the primaries (they specify the tri-chromatic gamut)
    -the gamma (or more generally the tonal reproduction curve or trasnfer
    curve)
    -the whitepoint

    So different RGB spaces behave differently, e.g. the same editing
    operations in different RGB spaces give different results.

    Why then do we have so many different RGB spaces? There are at least
    historical, practicality, marketing, political, religious (based on
    pure firm belief) and technical reasons for that.

    Timo Autiokari
     
    timo.autiokari, Jul 27, 2006
    #11
  12. ronviers

    ronviers Guest

    Thank you for your time and all the valuable information. I read your
    reply twice and got very sleepy. So I will sleep for a while then read
    it again then Wiki/Google some terms then read it again then look into
    the Photoshop tools you mentioned. Eventually I will get it.

    Brgds,
    Ron
     
    ronviers, Jul 27, 2006
    #12
  13. ronviers

    Mike Russell Guest

    The height will always increase when the tonal range is compressed, because
    there are more pixels at a particular value. It is not related to
    saturation. The min and max values of the histogram are often equated with
    saturation, but this is only a loose relationship. Conclusions based on
    interpreting a histogram, that are not also based on the appearance of the
    image, should be ignored.
    This is not surprising, the slider only affects the lightness channel, and
    is not hooked up to the a or b channels. The ability to alter Lightness
    without changing color is indeed a major advantage of Lab, and the reason I
    use it for the majority of my own images. If you bought Photoshop, you have
    paid for support for Lab, and it's worth your while to learn to use it.
    (The same goes for CMYK). But Lab is not the best choice for all images.

    In RGB, bumping the contrast of an image generally increases the color
    saturation. In practical terms, this makes RGB an ideal color space for
    dealing with underexposed images. Since such images are generally lacking in
    color as well as brightness, increasing contrast has the happy side effect
    of increasing color. The same is true of washed out images - making them
    darker by moving the black end of the RGB curve increases contrast, and adds
    saturation.

    Lab is my default color space for color correction work. If the saturation
    is already about right, RGB may add or remove saturation inappropriately.
    Most of my camera images benefit from a saturation boost, even though the
    exposure is correct. In addition to saturation, hue angle can also change
    as a side effect of any RGB correction. This is the main reason you may
    find yourself accidentally getting very saturated red or orange skin tones
    when correcting in RGB.

    HSB used to be directly supported in Photoshop, and the ability to
    manipulate saturation using a curve is very effective for certain images,
    such as landscapes. HSB has the interesting feature that increasing
    contrast can either add or remove saturation, depending on whether the
    contrast change makes the image darker or brighter.
     
    Mike Russell, Jul 31, 2006
    #13
  14. Please do this test:

    1) Take an underexposed raw shot of a color input target (e.g. the
    Kodak Q60). Use a strong underxposure, say -4EV, so that you will see
    the errors clearly.
    2) Convert the raw shot into a RGB color-space (but not to the sadRGB)
    and do not let the raw converter to correct the underexposure.
    3) Open the converted image to Photoshop.
    4) make a duplicate of that and convert the duplicate to Lab.
    5) Scale up the RGB mode image using the right input levels slider of
    the "RGB channel"--> the appearance of the image will pretty closely
    match the appearance of the target itself.
    6) Scale up the Lab mode image similarly using the right input levels
    slider of the "Lightness channel" --> the image shows _very_ larger
    luminance, hue and saturation errors.

    Please do the above experiment, you will learn that it is _not_
    possible to change the Lightness without changing "color" in the Lab
    mode.

    In this context the commonplace expression "color" is way too loose
    term. What we refer to with (reflected or self luminous) "color" is
    actually composed by three properties: Luminance, hue, and saturation.
    The luminance of a dark surface is low and the luminance of a white
    surface (e.g. illuminated by the sun) is high. The hue of the
    strawberry is red and the hue of the grass is green. Pink is less
    saturated red than what the saturation of the red of a strawberry is.

    A change (or scaling) on the "L"channel in the Lab space _does_ affect,
    very strongly indeed, to both the hue and the saturation, it is a
    widely spread misunderstanding that it would not. I have already
    presented a demonstration that shows that when the hue and/or
    saturation is changed in the Lab space then the luminance will change
    also. The contrary is _equally_ true, a change on the "L" channel will
    also change the hue and/or saturation of the color. Naturally the
    numerical values on the "a" and "b" channels do not change due to a
    change on the "L" channel but the displayed hue and/or saturation will,
    a great deal. Note that the "a" and "b" channels are not hue nor
    saturation -channels, they are just arbitrary numerical scales.

    The related operation in a RGB space is a change (or scaling) on the
    "RGB channel". The result depends on the properties of the particular
    RGB space. In case the RGB space is linear (has gamma 1.0 transfer
    function) or an accurate mathematical gamma function then the hue
    and/or saturation are not affected at all due to that operation and the
    lumimance scaling happens properly. In case of the sadRGB and the Lab
    spaces, hue and/or saturation errors are induced and the scaling of the
    luminance happens incorrectly.
    To what operation do you refer to by "bumping the contrast"? With
    digital images the "contrast" is somewhat a difficult property to be
    adjusted without creating some kind of errors, no matter in what color
    space you perform this such operation. In the real life it is much more
    easy to increase contrast, you just add more light to the scene. In
    digital imaging light can not be added freely since there is the upper
    limit (255,255,255), so other kind of rendering has to be performed in
    order to "bump up the contrast".

    I suspect the above could initiate some discussion about the property
    called contrast, so: Take absolutely all the light away from a scene,
    the result is that absolutely no contrast is available from the scene.
    Lit up a candle, now there is just a little of contrast available from
    the scene. Lit up a sun, plenty of contrast is available.
    Correctly chosen RGB space is ideal for all image manipulation.
    Underexposed images are just that, underexposed. They do not lack
    "color", nor saturation nor hue. They do not lack even the luminance.
    The digital coding of underexposed shots simply is scaled down due to
    the underexposure.
    Such adjustment moves the blackpoint of the coding upwards, this will
    _clip_ any data that there might be in that clipped portion of the dark
    end. In addition, if you do this in non-linear working-space then the
    "image-gamma" (a.k.a. "file-gamma") is affected resulting errors to all
    three color properties. Blackpoint adjustment should be performed only
    when the blackpoint is incorrectly coded.
    Editing work in a non-linear RGB working-space will create such (and
    other) errors, editing work in the Lab space also does so. Editing work
    in the linear RGB space either completely avoids or at least minimizes
    all such errors.

    Timo Autiokari
    http://www.aim-dtp.net
     
    timo.autiokari, Jul 31, 2006
    #14
  15. ronviers

    Mike Russell Guest

    ....
    I won't respond in detail to Timo's comments, if only because the topic has
    begun to stray from the original poster's concerns about changes in
    saturation due to interaction with other imaging operations.

    I would like to say that I believe Timo has contributed much careful thought
    and originality to what he is proposing here. People who are advanced
    enough in their knowledge of digital imaging would do well to examine what
    he says in detail. Timo has, at least in the past, been more than willing
    to respond individually to emails and I for one have found his ideas and
    discussions interesting and stimulating.

    As for using a linear RGB space, Timo is correct. Almost all, if not all of
    the algorithms used in graphics are designed for linear RGB, and there are
    color fringing and other artifacts introduced because of the fact that we
    work in a non-linear gamma space. I believe there is much food for thought
    in that one assertion alone, and that is why I singled it out for
    consideration.
     
    Mike Russell, Jul 31, 2006
    #15
  16. ronviers

    KatWoman Guest

    fwwwwooooooooomp
    sound of words flying right over my head.............................
     
    KatWoman, Aug 4, 2006
    #16
  17. ronviers

    Mike Russell Guest

    ....
    LOL - my apologies. With a little bit of luck, a cat (une jolie chat
    princess) can duck!
     
    Mike Russell, Aug 4, 2006
    #17
  18. ronviers

    KatWoman Guest

    splat cat on da floor

    I was an artist before any photo work (and came to that in the dinosaur time
    of film and paper, scraping silver off to retouch and using Doc martin dyes)

    I use PS in a completey intuitive way, by looking at the results by my eyes
    I know all you techies like the scientific methods and explanations.

    I dunno how my car works and I can't build one but I damn sure can put gas
    in and drive.
     
    KatWoman, Aug 8, 2006
    #18
  19. ronviers

    Mike Russell Guest

    [re tech speak and cabbages and kings]
    I'm a Spot Tone man myself.
    I think both are important, but if the rubber never meets the road, in terms
    of actual examples and images, the science can actually be worse than
    useless. Case in point - all the money currently being spent on profiling
    equipment, and the over-hype about color gamuts.
    That's all we ask, Katwoman. That, and more pictures. :)
     
    Mike Russell, Aug 8, 2006
    #19
  20. Unfortunately that alone does not make you a good, not to mention an
    excellent, driver, not in the normal traffic nor on the rally track.

    I often relate the task of image editing with that of a surgeon ... it
    is not enough that the surgeon knows how to hold and operate a knife.

    It is so with what ever task we perform, to know how to use a tool is
    just a very early beginning, in order to achive something good,
    something worth to mention, we do need to know a lot about the target
    itself and about the circumstances.

    Naturally one can learn some tricks or "recipes" that one can, more or
    less blindly, apply in Photoshop over an image after another but it is
    the same as driving a car without knowing anything about traffic
    regulations/rules/code and knowing nothing about the behavour of the
    car in some more demanding situations, so absolutely disastreous.

    In order to actually improve the images, to create something special,
    to invent new or to refine existing editing techniques, we simply must
    know, very deeply and in detail, the properites of those digital images
    that we are adjusting using the various tools and we need to know what
    are the properties that the tools affects to. And we need to know about
    the properties of the vision, the properties of various display and
    printing systems, about calibration and profiling. etc.

    Timo Autiokari
     
    timo.autiokari, Aug 8, 2006
    #20
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.