More LAB explorations

edgework

As a color retoucher, I, unlike most of you on this site, have no control over the images that i have to work on, and, truth is, despite the usual skin-tone, color cast and contrast issues, the images here are uniformly superior to anything I see in the professional ad world. Most look like they've come from the discount barrel at the local 99-cent store. I've found a LAB workflow to be invaluable in handling some of the disasters that cross my screen.

The group below is an example: the top image is what is called a "professional product shot", not something purchased from a stock art site but commissioned specifically for the ad. It was then scanned by a "professional" on a high-end drum scanner. It's virtually worthless as is. The image below it was worked entirely in LAB, and is a clear improvement.



So I'd come to think that everything I've known is obsolete, that LAB is the bright future, replacing all my old techniques. Wrong.

Consider this undeniably miserable image.



Anything that's good about it is hidden in the wash of a yellow cast, rotten contrast and blurred details. I thought it would be a perfect subject to test what LAB can do. Surely, I thought, LAB is the place to get rid of this cast and make everything beautiful.

I was mistaken. My LAB intensive version took many steps, wrong turns and detours, and then several additional moves in RGB and the results were never quite what I wanted.

So I started over, handling color and contrast in RGB before going into LAB, to see if I could improve the situation. Here's the LAB version, after I removed the cast in the highlights and shadows. A and B in both the white plate and the dark shadow in the middle of the stack of radishes in the back basket read 0,0. The bulk of the image is washed out from lack of contrast.



Here is my RGB version. My initial highlight reading, from the right side of the plate (for this trip, I'm ignoring the shadowy band along the left edge of the image) R=237, G=235, B=187. No surprise there. My shadow reading is picking up R=75, G=36, B=9. So yellow is too strong in the highlights and Red and Green are too weak in the shadows. My plan is to bring yellow into sync with Red and Green in the highlights, and darken the shadows in Red and Green to match Blue. Since all my target values are a safe distance from plugging up or blowing out, and there is serious need of contrast throughout, I simply set my anchor points to the desired values to get the steepest curve possible across the range.

These are the curves.



And this is the result:



Of course, the LAB version only dealt with the A & B channels, whereas the Red, Green and Blue curves steepened the contrast as well, but here was the surprise—I couldn't get a decent result using the lightness channel. I tried several variations, including fake black channel masks, and all manner of contorted curves. Here's as good a version as any:



The areas of main interest, the fruits and vegetables, seem to have a haze over them that I managed to overcome in later steps, but at a serious cost to other areas in the image. Now see the RGB version with an additional contrast move:



The detail is clearly crisper, the greens are far superior (note the light green on the outside of the lettuce wedges) and the interior of the lettuce wedge at 12 o'clock has much better shape and detail in it. (That area was the most difficult to solve with strictly LAB contrast moves.)

At this point I dumped my original LAB moves and took this new version into LAB and got the results I wanted in about half the steps.

The first LAB move is one that I haven't fully mastered yet, and it comes from Chapter 15 of Dan's book. It involves various blends of the A or B channel into the Lightness channel in Overlay mode. Since Neutral is always 50% gray, which has no effect in Overlay blends, the result is to darken and/or lighten certain colors to produce sometimes unexpected results that absolutely aren't possible in other spaces. While I understand the theory involved, I can't always envision what will happen, particularly if I use the inverse of A or B. So, as part of normal experimentation, I'll usually set up my Apply Image dialogue after selecting the lightness channel and setting the composite channel to view, and just see what happens. I'll try A, then its inverse, B, and its inverse. Sometimes nothing looks particularly good, or good enough to justify the move. In this case, the inverse of the B channel performed something of a miracle on the greens and yellows. Keep in mind that normally, Yellow is lighter, blue is darker in the B channel, so blending the inverse in overlay mode is going to have some kind of darkening effect on anything with yellow: reds, greens, and, of course, yellow. Here's what it looked like at 40% in the Apply Image dialogue:



The reds, most of the fruit and the darker areas of the greens have gotten unnecessarily heavy, but look at the detail in the lighter greens, and, particularly, the yellow hearts of the lettuce wedges. Much nicer detail.

Using these blending slider settings to eliminate reds and greens



produced this:



Next, I wanted to bring down the yellow that still is saturating the other colors. The fruits in the front bowl don't seem to have enough differentiation, the cauliflower pieces don't look white enough and neither do the cucumber slices, and the greens are still too yellow as well. And the corn, which should stand out as YELLOW seems lost in the mix.

This was a process of Command-clicking on different areas of similar color in an effort to drive them apart. I targeted various areas of red (orange fruit slices, tomato and pepper highlights and darker patches), greens and yellow (the lettuce heart at 12 o'clock, the center of the bowl and the corn. Also the cauliflower, which I pushed back towards neutral).

This is a procedure that I'm still studying; although I like the results I got here, I'm not sure I could explain exactly what I did. It was more a process of moving anchors up and down until things clicked. (One important point: unlike with MFM curves, the anchor points in this procedure should move Up and Down only.) As you can see by the curves



in all three color regions some parts have been pushed towards neutral. In this case, there really weren't colors that needed to be made a lot brighter, so the initial result appears to be a dulling of the overall color.



but yellow no longer dominates the scene. Now, a simple enhancement curve, pulling the yellow/blue anchors in 7 points and Red in 5 points (green was already nice and bright so I left it alone) seems to bring everything back, in terms of brightness, but without the yellow dominance. (Note now how the corn stand out.)

Now, Unsharp masking to the lightness channel (125, 1.0, 1) and an additional curve to a copy of the sharpened layer that opens up the parsely and greens on top (isolated with blending sliders and a soft mask), and the results are a definite improvement.



Bottom line: LAB is great for what it does, but it's not necessarily going replace techniques that have already proven themselves.

Tom K.

Thank you so much for this superb example. This is a real eye opener. It shows that one needs to keep all options on the table (pun intended). I will be studying this page very closely.

pathfinder

Did someone forget and shoot daylight film under tungsten light?? That is one yellow cast!!

edgework

pathfinder wrote:

Did someone forget and shoot daylight film under tungsten light?? That is one yellow cast!!

It's an extreme case, but not all that different from a lot of the stuff art directors give me.

Mark Horn

Thanks for your thoughts on this. I have been following all the threads related to the LAB Color with much interest. I find it hard to judge what will work with LAB and what won't. I must confess that having gone down the LAB path it is very addictive and hard to go back to RGB. I, as noted above, will be studying this posting very closely as it may answer some of the questions I have when the LAB editing doesn't go quite as planned. One thing I did note in the above example was your first attemp in LAB. The color seems good but the contrast is all wrong. What can work really well for this kind of haze is a USM (in RGB on the merged image) of USM 15/30/0 Repeat as many time as required. Its a bit like Hiraloam really but with slightly differnent settings. Obviously you can only repeat so many times before quality of the image goes out the window. Other setting are possible like USM 30/60/0 for a stronger effect but some times two lesser passes works better

edgework

Mark Horn wrote:

Thanks for your thoughts on this. I have been following all the threads related to the LAB Color with much interest. I find it hard to judge what will work with LAB and what won't. I must confess that having gone down the LAB path it is very addictive and hard to go back to RGB. I, as noted above, will be studying this posting very closely as it may answer some of the questions I have when the LAB editing doesn't go quite as planned. One thing I did note in the above example was your first attemp in LAB. The color seems good but the contrast is all wrong. What can work really well for this kind of haze is a USM (in RGB on the merged image) of USM 15/30/0 Repeat as many time as required. Its a bit like Hiraloam really but with slightly differnent settings. Obviously you can only repeat so many times before quality of the image goes out the window. Other setting are possible like USM 30/60/0 for a stronger effect but some times two lesser passes works better

I expected the LAB version to be lacking in contrast, since the cast removal only deals with the A and B channels. What I'm still working with is getting a comparable contrast result with the lightness channel.

Interesting approach with the sharpening. Usually I wait until everything else is done before sharpening, but I can see that your approach is more of a contrast move.

Don't get me wrong: I'm a confirmed LAB believer and Dan's techniques have changed the way I work. But it shows that, in LAB, we're not just translating familiar procedures into a new space; in many cases we're starting from scratch.

rutt

I had the chance to watch Dan at work just about a month ago. Even though his workflow now often passes through LAB, every image still has 10 channels and he blends channels and adjusts curves in all three color spaces. He insistes that image end up in CMYK for his class and this almost always requires/invites a final curve adjustment.

Dan is not teaching us to abandon RGB and CMYK by any means. Rather he is exploring the capabilities that LAB adds. I do think, though, that the typical DM workflow goes from RGB for cast reduction and contrast enhancing moves to LAB for color correction and sharpening (and fine tuning cast reduction and contrast ehancement) to CMYK for work with the deep shadows.

Mark Horn

Until I discovered the use of LAB I restricted everything to RGB as I only ever print to an inkjet printer. 99.9% of my learning has been via the internet and I've learn't a truck load but somewhere along the line I read that changing modes reduced the image quality - and hence limited my approach. Obviously this was a misconception or at least its something that happens in theory but in practice you can't detect the loss as long as the transition is through RGB along the way. I now will need to do some research on CMYK as I can honestly say I've never change modes to this.
Also have been following the CH16 thread. I've been able to get some great results using the techniques there and MFM's also. Using the hiraloam is great sometimes cause's a harsh result if skin not flawless like a childs. One way to overcome this is to use NeatImage or similar and crank it up a lot (to much if this was the final result - plastic skin etc) then bring it back into the original PS file as the top layer and tune the opacity to suit, perhaps with an edge mask to keep the detail sharp etc. Can even get pretty good results at ISO3200 doing this !

rutt

Mark Horn wrote:

Using the hiraloam is great sometimes cause's a harsh result if skin not flawless like a childs. One way to overcome this is to use NeatImage or similar and crank it up a lot (to much if this was the final result - plastic skin etc) then bring it back into the original PS file as the top layer and tune the opacity to suit, perhaps with an edge mask to keep the detail sharp etc. Can even get pretty good results at ISO3200 doing this !

The trick to making HIRALOAM less harsh is increasing the threshold and having a high enough radius. Try threshold values between 5 and 10. With a high enough radius, you shouldn't see it emphasize skin imperfections at all as the halos will be way too big.

For conventional USM, try doing it on a layer and using the blend-if sliders to eliminate the very positive flesh tones (postive A,B) from the blend. If that fails, also eliminate the lightest parts of the image, so you get only eyes and hair (sorry blondes). If you get frustrated you can always use a layer mask to eliminate just what you want. The effect of correct USM is subtle, so the layer mask will be easy to use.

If you want to actually improve the skin texture, particularly of women past a certain age, try using surface blur early on to close the pores and deemphasize the wrinkles. Do this on a layer so you can change opacity and blending options to target precisely, essentially the opposte of the USM targeting.

edgework

Mark Horn wrote:

I now will need to do some research on CMYK as I can honestly say I've never change modes to this.

When you move into CMYK there is great potential to degrade colors. The space is much smaller. Lots bright oranges, greens, blues and really intense reds just aren't possible (although you can get a purer yellow than RGB can). You'll need to acquaint yourself with the difference between Perceptual and Relative in the color settings dialogue. It refers to the way Photoshop maps colors from the larger RGB space to the more confined CMYK.

For example, if you have 10 shades of blue, and 6 of them are out of the CMYK gamut, a relative conversion will give you 10 relative shades, inside CMYK, even though none of them will actually match the original colors, even ones that were already in gamut. A perceptual conversion will convert colors that can be converted, and the rest will just get crunched. Sometimes intricate pattern detail, or shadow detail in blue draperies will turn to mud, and you have to resort to trickery to at least regain the shape of the original, if not the actual colors.

The number of situations where this is a serious problem is less than you would think, but you need to be aware of it, and, unless you're final destination is CMYK and you'll never need anything else, you should always keep an RGB master handy.

Another issue you'll need to worry about is the black plate. In theory, equal amounts of Magenta, Yellow and Cyan should create neutral, and 100% of each should give you black. It doesn't. For one thing, Cyan is a much weaker ink and requires about 10% more than Magenta and Yellow, in order to balance out a neutral tone. So C=100, M=100, Y=100 is going to give you a muddy brown. And, because there is a limit to the amount of ink you can lay down on a sheet of paper, it makes sense to siphon off the shadow tones into a single black ink channel, reducing the need for the other three inks. How that black plate is generated, how much you substitute black for the other tones and how soon you begin the substituition create wide variations in the way the colors are separated. The issues are many, and complicated, and you won't find a better treatment of them than Dan Margulis' "Professional Photoshop". I wouldn't even try to paraphrase the vast amount of information he offers up in that book, but it was originally conceived in the days when CMYK was the entire graphics industry, and it still retains a serious treatment of how to make that space perform at its best, and how to make it get along with the other types of output requirements that have evolved since it was first written.

Mark Horn

rutt wrote:

The trick to making HIRALOAM less harsh is increasing the threshold and having a high enough radius. Try threshold values between 5 and 10. With a high enough radius, you shouldn't see it emphasize skin imperfections at all as the halos will be way too big.

For conventional USM, try doing it on a layer and using the blend-if sliders to eliminate the very positive flesh tones (postive A,B) from the blend. If that fails, also eliminate the lightest parts of the image, so you get only eyes and hair (sorry blondes). If you get frustrated you can always use a layer mask to eliminate just what you want. The effect of correct USM is subtle, so the layer mask will be easy to use.

If you want to actually improve the skin texture, particularly of women past a certain age, try using surface blur early on to close the pores and deemphasize the wrinkles. Do this on a layer so you can change opacity and blending options to target precisely, essentially the opposte of the USM targeting.

Thanks...........I have been using the threshold but sometimes I want more definition in the cheek bones and as the threshold increases the effect is just a little subtle. Blend-if sliders also a good alternative. Much more effective (ie better transitions even with the slider split) in LAB than RGB I think or at least easier to get a good result.

Thanks for the Darken/Lighten thread ......... excellent stuff that one.

Mark

Mark Horn

edgework wrote:

When you move into CMYK there is great potential to degrade colors. The space is much smaller. Lots bright oranges, greens, blues and really intense reds just aren't possible (although you can get a purer yellow than RGB can). You'll need to acquaint yourself with the difference between Perceptual and Relative.................

Thanks.

Mark