LAB color space not ideal for color correction?

glennchan

Is LAB ideal (or not) for color correction? It does have some interesting properties. However, I'd argue that it doesn't have the right kind of perceptual uniformity that is useful for image manipulation / color correction. Almost every other color space does a better job at maintaining constant hue when you manipulate saturation. You can read my full blog post here, with an example image:
http://colorcorrection.info/?p=13

In the same vein, LAB color space is prone to issues that chromaticity based color spaces face. Sometimes manipulations to the Lightness channel causes unintended shifts in saturation.
http://colorcorrection.info/?p=15

My apoligies to Dan Margulis. I don't mean to stir the pot. Well ok, maybe a little bit.... hopefully this gets you thinking about different color spaces!

Richard

glennchan wrote:

Is LAB ideal (or not) for color correction? It does have some interesting properties. However, I'd argue that it doesn't have the right kind of perceptual uniformity that is useful for image manipulation / color correction. Almost every other color space does a better job at maintaining constant hue when you manipulate saturation. You can read my full blog post here, with an example image:
http://colorcorrection.info/?p=13

In the same vein, LAB color space is prone to issues that chromaticity based color spaces face. Sometimes manipulations to the Lightness channel causes unintended shifts in saturation.
http://colorcorrection.info/?p=15

My apoligies to Dan Margulis. I don't mean to stir the pot. Well ok, maybe a little bit.... hopefully this gets you thinking about different color spaces!

Margulis has quite a few fans here and a little opposition as well, so it's possible that the pot will boil over once again. FWIW, I use Dan's techniques quite often as they are both fast and powerful, but I also work in RGB and CYMK when those give me an easier path to my goal. I also have little doubt that other color spaces have advantages in certain cases, but I just haven't needed them yet. I guess what I am trying to say is that I don't really believe that there is an "ideal" color space any more than there is an "ideal" selection tool in Photoshop. The best choice will depend on the image, the goal, the output medium, the time available and the skill of the user.

arodney

One of the best quotes about Lab comes from the great, late Bruce Fraser:

Let me make it clear that I'm not adamantly opposed to Lab workflows. If
they work for you, that's great, and you should continue to use them.

My concern is that Lab has been oversold, and that naive users attribute to
it an objective correctness that it does not deserve.

Even if we discount the issue of quantization errors going from device space
to Lab and vice versa, which could be solved by capturing some larger number
of bits than we commonly do now, (though probably more than 48 bits would be
required), it's important to realise that CIE colorimetry in general, and
Lab in particular, have significant limitations as tools for managing color
appearance, particularly in complex situations like photographic imagery.

CIE colorimetry is a reliable tool for predicting whether two given solid
colors will match when viewed in very precisely defined conditions. It is
not, and was never intended to be, a tool for predicting how those two
colors will actually appear to the observer. Rather, the express design goal
for CIELab was to provide a color space for the specification of color
differences. Anyone who has really compared color appearances under
controlled viewing conditions with delta-e values will tell you that it
works better in some areas of hue space than others.

When we deal with imagery, rather than matching plastics or paint swatches,
a whole host of perceptual phenomena come into play that Lab simply ignores.

Simultaneous contrast, for example, is a cluster of phenomena that cause the
same color under the same illuminant to appear differently depending on the
background color against which it is viewed. When we're working with
color-critical imagery like fashion or cosmetics, we have to address this
phenomenon if we want the image to produce the desired result -- a sale --
and Lab can't help us with that.

Lab assumes that hue and luminance can be treated separately -- it assumes
that hue can be specified by a wavelength of monochromatic light -- but
numerous experimental results indicate that this is not the case. For
example, Purdy's 1931 experiments indicate that to match the hue of 650nm
monochromatic light at a given luminance would require a 620nm light at
one-tenth of that luminance. Lab can't help us with that. (This phenomenon
is known as the Bezold-Brucke effect.)

Lab assumes that hue and chroma can be treated separately, but again,
numerous experimental results indicate that our perception of hue varies
with color purity. Mixing white light with a monochromatic light does not
produce a constant hue, but Lab assumes it does -- this is particularly
noticable in Lab modelling of blues, and is the source of the blue-purple
shift.

There are a whole slew of other perceptual effects that Lab ignores, but
that those of us who work with imagery have to grapple with every day if our
work is to produce the desired results.

So while Lab is useful for predicting the degree to which two sets of
tristimulus values will match under very precisely defined conditions that
never occur in natural images, it is not anywhere close to being an adequate
model of human color perception. It works reasonably well as a reference
space for colorimetrically defining device spaces, but as a space for image
editing, it has some important shortcomings.

One of the properties of LCH that you tout as an advantage -- that it avoids
hue shifts when changing lightness -- is actually at odds with the way our
eyes really work. Hues shift with both lightness and chroma in our
perception, but not in LCH.

None of this is to say that working in Lab or editing in LCH is inherently
bad. But given the many shortcomings of Lab, and given the limited bit depth
we generally have available, Lab is no better than, and in many cases can be
worse than, a colorimetrically-specified device space, or a colorimetrically
defined abstract space based on real or imaginary primaries.

For archival work, you will always want to preserve the original capture
data, along with the best definition you can muster of the space of the
device that did the capturing. Saving the data as Lab will inevitably
degrade it with any capture device that is currently available. For some
applications, the advantages of working in Lab, with or without an LCH
interface, will outweigh the disadvantages, but for a great many
applications, they will not. Any time you attempt to render image data on a
device, you need to perform a conversion, whether you're displaying Lab on
an RGB monitor, printing Lab to a CMYK press, displaying scanner RGB on an
RGB monitor, displaying CMYK on an RGB monitor, printing scanner RGB to a
CMYK press, etc.

Generally speaking, you'll need to do at least one conversion, from input
space to output space. If you use Lab, you need to do at least two
conversions, one from input space to Lab, one from Lab to output space. In
practice, we often end up doing two conversions anyway, because device
spaces have their own shortcomings as editing spaces since they're generally
non-linear.

The only real advantage Lab offers over tagged RGB is that you don't need to
send a profile with the image. (You do, however, need to know whether it's
D50 or D65 or some other illuminant, and you need to realise that Lab (LH)
isn't the same thing as Lab.) In some workflows, that may be a key
advantage. In many, though, it's a wash.

One thing is certain. When you work in tagged high-bit RGB, you know that
you're working with all the data your capture device could produce. When you
work in Lab, you know that you've already discarded some of that data.

Bruce

Its been way over sold, especially if you're working with Raw files in a good Raw processor. Using those tools should (and I would love to see the Lab proponents prove otherwise) diminish the need for Lab to about nill. Its a non intuitive space to work in numerically. There's data loss and time lost going from space to space. It brings little if anything to the party that a good RGB working space can accomplish after proper use of RGB data in good Raw processor.

jjbong

glennchan wrote:

Is LAB ideal (or not) for color correction? It does have some interesting properties.

Your two blog posts are quite clear and well illustrated. However, the question you raised (is LAB ideal [or not] for color correction) is a very large one. The issues for someone developing a color correcting plug-in are very different from the issues for someone using PhotoShop to do color correction (such as myself). I suspect your post is primarily directed toward the former.

I use LAB all the time for specific objectives in the process of color correction, but I never use it for the entire job (or even a large part of it). In particular, I would never use it to do either of the things you illustrate (reducing saturation and reducing overall brightness), or even use an S-curve on the L channel to enhance contrast (which you mention).

Here's where I find LAB to be particularly useful:

1. Limiting corrections (often made in RGB) to particular areas of the picture, as in many cases they can be differentiated successfully by their A and B components.
2. Generating masks (L channel frequently, but also A and B occasionally).
3. Adding a bit of pop after the rest of the correction/enhancement has been done.
4. Sharpening.

I'm not saying that LAB is the only way to do this or even the best way (I can't claim the expertise or experience to make such a statement). I'm just saying that it's worked for me in these cases, it's relatively fast, and I use it all the time.

I also find it useful to have the Info palette display the LAB values when I'm working in RGB or CMYK. The most useful thing is that it shows you quickly when you have a subtle cast (green or blue hair, in the circles where I travel, is dead wrong). As many others have pointed out, getting casts out quickly makes the whole picture better.

There are common cases where LAB won't work easily, if at all. In addition to the issues you raised, I'll add one that I see all of the time. Color casts that vary over the image (usually, but not always) by luminosity. RGB curves work fine for this.

Like Richard, I occasionally use CMYK, mainly for generating a K channel for use in RGB, but also for curving close-ups of people (and doing quick "touch ups" of reddish spots on faces with channel blending).

Richard wrote:

I guess what I am trying to say is that I don't really believe that there is an "ideal" color space any more than there is an "ideal" selection tool in Photoshop. The best choice will depend on the image, the goal, the output medium, the time available and the skill of the user.

This is the most sensible summary I've read.

glennchan

As far as generating masks go, I think it would be useful if Photoshop were able to isolate colors based on some chromaticity based system.

Suppose you make a gradient from red to black. In LAB space, the a and b channels will be a gradient and not the same value. Whereas in HSB space (e.g. check this via the eyedropper), the H and S channels are uniform throughout (except at black?).
So if there was a red object in a real world scene with varying illumination on it, it would be close to the gradient example. In a space like HSB (or some other chromaticity based color space), you would only need to select a smaller range of color compared to a chrominance based color space.

S-curve on the L channel to enhance contrast

I believe it's one of the techniques advocated in Dan Margulis' book.

2- This is not to say that you can't do good work in LAB space... but I think there are some other color spaces that would be slightly better in the same situations.

jjbong

glennchan wrote:

As far as generating masks go, I think it would be useful if Photoshop were able to isolate colors based on some chromaticity based system.

No argument here. Some of the tools in Photoshop are incredibly crude, as they haven't been advanced in a very long time. Selective Color, for just one other example, has a very limited way to select color ranges. And there's no meaningful way to work in HSB, as near as I can tell.

Some of these discussions get unnecessarily heated because two different topics are debated as if they were one: what would be a superior approach if it were available in PS, and what's a good way to solve a problem with the tools currently available.

glennchan wrote:

I believe it's (S curve on L Channel) one of the techniques advocated in Dan Margulis' book.

Possibly (I don't remember). It certainly doesn't seem to be his current thinking, based on his Picture Postcard Workflow. His contrast enhancement in this workflow is based on channel blending and curves in RGB, with a few other things thrown in (mostly RGB).

arodney

jjbong wrote:

Possibly (I don't remember). It certainly doesn't seem to be his current thinking, based on his Picture Postcard Workflow. His contrast enhancement in this workflow is based on channel blending and curves in RGB, with a few other things thrown in (mostly RGB).

But never at acquisition (scanning or Raw conversion) which I find so very, very odd.

BinaryFx

Some links to other discussions on Lab editing:

http://www.alternativephotoshop.com/phpBB3/viewforum.php?f=7&sid=0f1665d3b78ca970902a433258e8f5d9 (colour correction forum)

http://www.ledet.com/margulis/ACT_postings/ColorCorrection/ACT-LAB_intuitive.htm

http://www.ledet.com/margulis/2007HTM/ACT06-LAB_opposing_color.htm

http://www.ledet.com/margulis/ACT_postings/ColorCorrection/ACT-LAB-damage.htm

http://www.ledet.com/margulis/2007HTM/ACT06-LAB_Conversions.htm

http://www.ledet.com/margulis/ACT_postings/ColorCorrection/ColorCorrection.htm


Hope this helps,

Stephen Marsh
http://members.ozemail.com.au/~binaryfx/