Chapter 2, Professional Photoshop, 5th Edition

Andy · December 30, 2006

Chapter 2 - The Steeper the Curve, The More the Contrast

Dan Margulis wrote:

Writing curves is a form of horsetrading. Most improvements in the image come at a price. Fortunately, the price is sometimes quite reasonable. Look for areas you wish to emphasize, and then figure out how you will pay for that improvement.

Dan hits us over the head again with the orientation of the Curve - switch it so that the dark parts are upper and right. In CS2, you click on the gradient. In CS3, you hit the down arrow to show more details, and you select "pigment inks." If you do this, you'll be in sync with the curves illustrations in the rest of the book, and in all the reading group summaries. There's no sense really debating it - either you do it or you don't, it's entirely up to you

It's all About Contrast

When we view an image, our eyes are craving contrast. Images with good contrast are pleasing to us. Images with little or poor contrast, are typically described as "flat" "grey" "dull" or similar. This is not to say that you must have gobs of contrast in all images! But for the most part, most photos will benefit from it. Here on Dgrin, we've been calling it "Pop" and we have several good tutorials that will get the new or learning Photoshopper on the way to using curves for better contrast.

Setting the Black and White Point

Dan doesn't talk about this much in this chapter, but it's what I usually do first. Here is an excellent method of finding the important black and white points in your image. Remember, the chosen points may not be the darkest and lightest. Dan spends some time on "avoiding the catchlights" or what others call, the specular highlights. This is another reason why shooting in RAW, where we as photographers get to choose what's important, not some algorithm in the camera jpg, or in Photoshop's Auto Levels dialog. Why? If you set the endpoints to their very extremes (0 and 255 for example), you are for sure increasing contrast, but losing important detail in the highlights or shadows. How important that detail is, will vary by image, of course.

[imgr]http://www.moonriverphotography.com/photos/119834978-S.jpg[/imgr][imgr]http://www.moonriverphotography.com/photos/119834946-S.jpg[/imgr]

Curve Shape

Is an overall big ole S-curve always right? Well, it typically won't hurt, but it's not always best. The S-Curve will generally steepen (add contrast) to the midtones of the image. But what if the important areas are in the shadows or highlights? The typical S-curve does not improve these areas.

Curve Point Selection, and moving the curve. Click and hold, move your cursor around the image and when you have a point you want to plot on the curve, command-click it. NOW, when you want to move it, do as Dan says, and move it North-South only - use CTRL-TAB to move between the plotted points and the up/down arrows to now move (steepen) the curve. Thanks DavidTO, I've been seriously lagged in this department. Very precise, very easy.

Individual RGB vs. Master Curves

Margulis says that you'll often get better results by curving the channels individually. We'll see in later chapters. Dan says to avoid the master curve, that in CMYK it's a disaster, because K does not behave like CMY. OK fine, but we are photographers, and for most of us, we'll work in, and stay in, RGB curves. Still, Dan asserts that most images can benefit from individual moves in each channel's curve, rather than a master curve. I'm looking forward to having this lightbulb go off for me, personally, because I haven't yet found the pay-off in doing these curves individually (each channel separately). I'd love you to show me how!

So, in summary: Most images will benefit from added contrast. Use the curves tool to achieve this contrast. Taking control away from your camera (shooting RAW, or adding as little contrast in your in-camera jpgs as possible), and deciding what the important areas of the image are - this is what this chapter is about. Oh and don't forget the benefits of using the adjustment layers. You must get this concept now or you'll be hosed later. If you need to dial an adjustment back some, well, it's a cinch with the opacity of the adjustment layer! We'll use some of the techniques and fundamentals learned here throughout the book, and I'm counting on Dan to get that light bulb to turn on for me, with regard to "individual r-g-b curves will usually yield a better result."

The other thing I struggle with is that Dan's big on "global moves" and well, that's fine but I won't stick with global moves if there's a 60 second local correction or improvement that I can make. It's probably just that I'm accustomed to some ways that I've been using, and haven't yet gr0k'd the use of curves for a "local" improvement. I suspect this has to do with more sophisticated moves one will make with the curve, in different channels.

Some fun examples for us to play with in this chapter discussion thread. I'd love to see what you guys can do with them. They are unprocessed, except for converting from RAW. Click the thumbnails to get the original files.

System · December 30, 2006

I'm always wondering if this guy was crackin' wise with you when you took his pic-

adjusted curves separately in r, g and b channels and also the rgb-

DavidTO · December 30, 2006

I'd love to see someone (especially one of the color theory group guys) explain this whole adjusting the curves separately, and show how it's done, getting a successful version of one of these images. I gave up on it. It's beyond me at this point.

Anyone?

Andy · December 30, 2006

gefillmore wrote:

I'm always wondering if this guy was crackin' wise with you when you took his pic-

adjusted curves separately in r, g and b channels and also the rgb-

I think you went too far. Lots of details lost, hair and shadows. And you made colors there that shouldn't be there (greens). This is my point about the r-g-b curves separate adjustments, how do you know? I'm not picking on you, George ...

This was my edit, just one master curve adjustment. I'm still waiting for an r-g-b curve that's better. Thanks George for contributing!

System · December 30, 2006

no problem-

I don't get too concerned about shadows, but maybe I should-

and yeah, it's green-

back to the drawing board?-

is this where the adjustment layers come in, where you can adjust for colors and highlights and shadows?-

Andy · December 30, 2006

gefillmore wrote:

is this where the adjustment layers come in, where you can adjust for colors and highlights and shadows?-

Well, according to Dan, we shouldn't have to do anything but global moves. Where' I'm stuck is in why/how r-g-b moves are better/easier than a single master curve move (which I did in my edit just above).

System · December 30, 2006

well, I eagerly anticipate the answer to this 'conundrum'-

Richard · December 30, 2006

Andy wrote:

Well, according to Dan, we shouldn't have to do anything but global moves. Where' I'm stuck is in why/how r-g-b moves are better/easier than a single master curve move (which I did in my edit just above).

Yeah, I am stuck at the same point. I understand his theoretical point--that master curve changes in RGB can distort colors--but since I don't have nearly the grasp of channels that Dan has, I generally seem to screw up the colors more when I try to muck with the individual channels. I do most of my work in LAB, though, where it is easier to separate color and contrast moves.

jfriend · December 30, 2006

Separate color from contrast

Andy wrote:

Well, according to Dan, we shouldn't have to do anything but global moves. Where' I'm stuck is in why/how r-g-b moves are better/easier than a single master curve move (which I did in my edit just above).

I have played with per-channel moves and also can't get them to work the way he describes. The thing I don't understand about them is that unless you make the exact same moves in each channel (and then, what's the point), you are going to create significant color shifts in some tonal ranges. I find it way, way, way easier to do as Dan taught us in the LAB book and separate color from contrast. Make a contrast move with either an RGB curve in luminosity blend mode or an L curve in LAB mode and then make separate moves to adjust the colors either with R, G or B curve moves or A or B moves in LAB. In my normal workflow, I first get the global white balance right (either in ACR or with R, G or B curve moves), then enhance contrast (with a global curve), then decide if any finer grained color moves are needed (by shaping individual color channels).

When you are curving the R, G and B curves and trying to adjust color and contrast at the same time, it's either something only the Gods can do or it's just impractical and not nearly as easy as separating contrast moves from color moves like Dan taught us in the LAB book.

edgework · December 30, 2006

Andy wrote:

Well, according to Dan, we shouldn't have to do anything but global moves. Where' I'm stuck is in why/how r-g-b moves are better/easier than a single master curve move (which I did in my edit just above).

In theory, because the three RGB channels are of equal strength and equal values in all three yields dead neutral, master curves should simply darken and and lighten without shifting hue or saturation. It doesn't work that way. Darkening the Master curve pushes towards red. Setting a curve adjustment layer to Luminosity mode can ease that problem some, but there is still a red shift. Try it by taking a copy of the same image into LAB and pushing the lightness curve. Comparable moves in RGB will not produce identical results.

As for single channel moves, the point is to stretch the tonal range of each channel to maximize the full spectrum. Unless you are certain that the three channels are strong and deficient in the same ranges, a single, one-size-fits-all curve won't really fit any individual channel. The blue channel might need a boost in the shadow range while the red could need highlights stretched. No single Master curve can accomplish both.

As for global moves, I've always looked at it as Strategy vs. Tactics. Global moves are a sound strategy; if magenta is deficient in one part of the image, it's probably out of whack all over. But any general will agree that tactical requirements in the field trump stragegic planning every time. I like to work globally as much as I can, but sometimes a face needs different treatment. If the face is the point of the image, whether or not it balances perfectly with the background may not mean much. Ya gotta call 'em as you see 'em.

DavidTO · December 30, 2006

edgework wrote:

As for single channel moves, the point is to stretch the tonal range of each channel to maximize the full spectrum. Unless you are certain that the three channels are strong and deficient in the same ranges, a single, one-size-fits-all curve won't really fit any individual channel. The blue channel might need a boost in the shadow range while the red could need highlights stretched. No single Master curve can accomplish both.

Cool, I'd love to see it on that shot Andy has above of the man smiling at camera. The one the gefillmore took a whack at. I need to see someone do a channel by channel correction of that image that looks good. Like I said, I've given up.

pathfinder · December 30, 2006

I am so pleased to learn that I am not the only soul who has difficulty achieving good color and contrast balance by adjusting individual R,G, or B channels. I understand the theory, I think. I just find it rather hard to grok it out.

If I can find a black or a white or a good neutral, I can usually be confident of my results. But if they are not present, it gets much harder for me.

I too am more comfortable in LAB than RGB, and least comfortable in CMYK.

Here is my try at Andy's dog

And his tailor

Not much editing needed by either of these shots though.

kwalsh · December 31, 2006

OK, so I went for the "What's the difference between composite and separate channels?" route.

Starting assumption: It's the guy's mug we care about. We want it to "snap" and so we want maximum contrast on his face.

Method: I don't want to dork with color, so it is an adjustment layer set to luminosity mode. Somewhere in the first few chapters Dan mentions many of his students do curves twice, once for contrast and once for color using the luminosity and color blend modes. Since chapter 2 is just on contrast I'm using the luminosity blend. Side note: If you think a composite curve doesn't effect color go ahead and click "luminosity" and be surprised just how much it changed the color!

So, set white and black points by moving the top and bottom of the curve and check "show clipping" and wait for something important to clip. In composite mode you can't do much, in per channel you can move 'em a lot.

Next, examine the face with the dropper and see where it lives in the curves. In composite it lives mostly above the 1/4 tone and is spread all over the place. In per channel it is in the middle for blue and near the middle top for green and very near the top for red. So blue will be an S and red and green just exponentials.

A word of apology, I forgot to flip the curves into "Dan" mode so they are backwards for all ya.

Final Results: I got a fair bit of contrast with both - but I couldn't get the composite curve to give me as much as the per channel. It always ended up either biased light, dark or if I did an S lost a lot of contrast. What I've posted here was about as much contrast as I could and ended up a bit light compared to the per channel.

Here are the curves:

Here is the full frame with per channel curves:

Here it is with composite:

And since the face is what I'm intersted in, left is per channel, right is composite:

So here it seems to me the face does have higher contrast while holding detail in the shadows and highlights. Unfortunately I couldn't get the composite to exactly match the brightness of the per channel without it going flatter on me. I'm not sure if that's a function of the difficulty of a composite channel or me just not being good enough.

So, in conclusion, I'm starting to get what Dan means here. I just don't think I'm clever enough to do contrast and color simultaneously and it sounds like lots of his students don't either! Are the differences here super drammatic? No, but I'm not sure much of this is (short of the LAB tricks that is). Anyway, hopefully this sort of illustrates using per channel curves, even if it is my inept first attempt.

Ken

jfriend · December 31, 2006

Color channel moves with luminosity blend mode are in the LAB book

kwalsh wrote:

So, in conclusion, I'm starting to get what Dan means here. I just don't think I'm clever enough to do contrast and color simultaneously and it sounds like lots of his students don't either! Are the differences here super drammatic? No, but I'm not sure much of this is (short of the LAB tricks that is). Anyway, hopefully this sort of illustrates using per channel curves, even if it is my inept first attempt.

In the LAB book (I don't remember which chapter), Dan does suggest exactly the technique you used here - adjusting each R,G,B curve separately in the luminosity blend mode. I've found it to be a useful technique for contrast enhancement on some particular types images. In fact, it's even better than an L curve in LAB mode on some images.

The best I can tell, the images that it works uniquely well on are ones where the R, G and B channels all peak in very different locations, thus you can't target a single area of the composite curve to steepen and not miss a lot of interesting pixels, some of which you care about. Furthermore, when one of the areas you're trying to enhance contrast is pretty bright or pretty dark, it seems like it's easier to steepen a particular color channel in that region without throwing off the rest of the image as much as doing so in the composite channel causes.

From what I've read (and remembered) in this new book, I don't see Dan recommend using the Luminosity blend mode with his color channel moves. He seems to just have at it directly on the color channels, somehow doing simultaneous moves in color and contrast. I'm sticking to one at a time. Much more predicatable, quicker and more reliable for me. I don't think there are any disadvantages to separating the color and contrast moves either.

kwalsh · December 31, 2006

jfriend wrote:

The best I can tell, the images that it works uniquely well on are ones where the R, G and B channels all peak in very different locations, thus you can't target a single area of the composite curve to steepen and not miss a lot of interesting pixels, some of which you care about. Furthermore, when one of the areas you're trying to enhance contrast is pretty bright or pretty dark, it seems like it's easier to steepen a particular color channel in that region without throwing off the rest of the image as much as doing so in the composite channel causes.

Yes, this is definately true I think. I don't think the doggie in the snow is going to have any help from per channel curves. Human faces on the other hand seem to be a reasonable target. I think accordian man will be problematic since we need to preserve the accordian - perhaps that'll be a CMYK with the K curve for the accordian. Yosemite belongs in LAB for sure.

One visualization trick if the whole mouse over the subject and watch where it lies on the curve thing doesn't do it for you. On the adjustment layer do each channel one at a time. Before doing the curve for a channel go to the channels panel and show only that channel. That way you'll basically be doing curves for three separate B&W images. Might be easier to digest at first.

Ken

P.S. I need to read the LAB book again too...

rutt · December 31, 2006

I'm in Italy until middle of next week and have only the most unreliable internet connection (thanks to tin.it)

Hey, Andy, good job. That's what I like to see, explain what you get, ask questions about what you don't get. Provoke discussion.

I've come to see RGB curves as an advanced technique. LAB curves are much easier. The L curve does just what you expect for contrast and the A and B curves make big color moves pretty easy. Combine with blend-if sliders and you get a lot of control pretty easily.

If you do need to work in CMYK, you just have to use the curves to adjust the shadows by steepening the dark end of the K curve and removing some of the other colors from the shadows (yellow in particular.)

RGB curves, on the other hand, are a less important tool in my box. My most frequent use of them is to adjust skin tones. I just did one, but I don't hink I can post. But when flesh is more magenta than yellow, make a curves layer, pull up the blue curve, find out where the magenta flesh tones live on the curve and pull the curve up just a little there (toward darkness). By darkening the blue, you are lightening the yellow and bringing it back into balance. Because you only change the curve where the fleshtones are, you have less chance of changing neutral points or the color in other parts of the image. Once you are done getting the flesh right, you can set the blending mode of the curve layer to "Color" and it won't effect contrast. Of course, since you have only played with the blue curve, it will have only a 10% effect on contrast, right?

Dan has plenty of examples with individual channels in RGB curves being particularly important. I'll scan through the book and cite some of them.

The white dog is like the most classic example of something I would do in LAB. The first class of Dan's we did an image of some white seals. We did it three times, once with CMYK curves, once with blending, and finally with LAB curves. The point was that LAB curves and sharpening did the best job easiest by far of keeping the neutral neutral.

Andy · December 31, 2006

Looking at Ken's and PF's edit, with r-g-b individual curves, I don't see an improvement over the master curve adjustment. I see loss of detail, too much contrast, and too much saturation.

I'm going to continue to be patient though, and wait for it to hit me like a ton of bricks

kwalsh · December 31, 2006

Andy wrote:

Looking at Ken's and PF's edit, with r-g-b individual curves, I don't see an improvement over the master curve adjustment. I see loss of detail, too much contrast, and too much saturation.

I'm going to continue to be patient though, and wait for it to hit me like a ton of bricks

Well I don't think there is any loss of detail in either the composite or per channel edits on my two images. Actually the detail is greatly enhanced because of the (excessively?) enhanced contrast.

I'm totally with you on the no huge difference between the two. Of course a lot of the differences between images in the book are pretty subtle as well. I too have yet to be hit with a ton of bricks

.

As far as too much contrast goes, I'm not going to argue with you there. I was just taking Dan's "emphasize contrast on the subject" concept to the maximum. I'm not sure we really want to do that to this poor guy!

Taking another card from Dan's deck and over doing something and then dialing it back here is the per channel with the opacity dialed back to 50%:

I greatly prefer this one to any of the previous posts. My ultra-high contrast ones are just too dark. Comparing this version to your's and PF's the face has much more depth and shape to it and he is set off from the background better. That said, too each their own! And now with it dialed back to 50% it would seem there would be even less difference between this and a composite curve. Probably need some more dramatically colored image to see a big difference.

Anyway, still playing and reading and looking around expectantly for bricks...

Ken

P.S. Great image, by the way.

edgework · December 31, 2006

Aside from an overall lightness in the shadows, the entire image feels too red. On first reading, the skin tones seem okay from the numbers. The biggest problem is that the face is flat. The only real variation in detail comes from the blue plate, which includes a host of skin imperfections that makes it unacceptable for use in plate blending.

The shadows in the clothing are weak, but also red; ideally, you would push towards neutral as you darken in areas. The most telling feature is his hair. Maybe he's still got light brown hair, but I'd bet that he's closer to grey than the numbers suggest: 36C 37M 43Y doesn't suggest light brown to me so much as a red cast.

The curves I used push the cyan element with a larger bulge in the Red channel, and pull all three down in the highlights, to bring out more detail in the face. I focused on different areas of the face which, predictably, has a variety of tones and hues, tweaking curves in the Green and Blue channels to keep either magenta or yellow from dominating. I added cyan by pushing the Red curve in the shadow areas in the clothing, and in the dark lines of his face, which you want to keep from going hot. Most of those shadows were in the cyan midtones, with the exception of the dark suit to the right of the image.

As it happens, without actually measuring any of the hair, it now looks a good deal grayer. Also, the shadows have pushed more towards neutral without losing the brillance of objects that are truly red, green and yellow. Note also, the increased detail in the red plaid at the left of the image.

(Since this is about curves, I avoided the big gun in face balancing: Selective color, with a slight subtraction of Magenta and addition of Yellow to the Reds, and a subtraction of Yellow and addition of Magenta to the Yellows).

And, following the "Rule-of-75," which states that any move that looks good upon completion, will look much better the next day if you first pull the change back to 75% of its original intensity, we get this version:

Another move I avoided, but which I ordinarily always use, is to make a fake black adjustment by copying the image, converting to CMYK and then using the black channel as a layer maske for a luminosity curve (the master curve works fine for this), adding weight to shadows in a much more subtle way than a raw Master curve could accomplish.

These are the three curves I used:

Duffy Pratt · December 31, 2006

I own Samoyeds myself, so I ordinarily would have done the Samoyeds in LAB. For this, I did it in RGB, partially because I thought there was a slightly moving cast on the dog. I decided to neutralize the bluest parts of the dog, and let the snow go even more blue. I also picked the top of the dogs nose as the shadow, but set it's black point to a very conservative 30,30,30. For the white point I took the lightest point on the nose and set it to 245,245,245. This adjustment here is only the RGB curve, done by individual channels.

For the smiling man, I did a CMYK curve. The black point was basically a guess, and I was willing to be flexible on it. I decided on the sleeve of the jacket behind the window. The white point was the lightest part of the man's hair. The biggest problem with this picture was the excess magenta in the skin, and the big color variations in the skin as well. Again, this is just a CMYK curve and nothing else.

On the general issue, its been important for me to write curves on the individual channels even in the easy cases where a master curve could do, largely for practice and to learn. When things go haywire, I can always retreat to easier methods, but I find that things go haywire less often the more I try to use the methods Margulis recommends. And overall, the results seem to be better for me.

Duffy

Duffy Pratt · December 31, 2006

Can only upload one file at a time, it seems. This is the tailor done in individual CMYK curves, as discussed above.

Duffy

DavidTO · December 31, 2006

Duffy Pratt wrote:

Can only upload one file at a time, it seems. This is the tailor done in individual CMYK curves, as discussed above.

Duffy

You can use IMG tags to put more than one pic in a post, but yes, attach is one at a time.

Check out the link in my sig for a how-to.

jfriend · December 31, 2006

rutt wrote:

RGB curves, on the other hand, are a less important tool in my box. My most frequent use of them is to adjust skin tones. I just did one, but I don't hink I can post. But when flesh is more magenta than yellow, make a curves layer, pull up the blue curve, find out where the magenta flesh tones live on the curve and pull the curve up just a little there (toward darkness). By darkening the blue, you are lightening the yellow and bringing it back into balance. Because you only change the curve where the fleshtones are, you have less chance of changing neutral points or the color in other parts of the image. Once you are done getting the flesh right, you can set the blending mode of the curve layer to "Color" and it won't effect contrast. Of course, since you have only played with the blue curve, it will have only a 10% effect on contrast, right?

I agree that I use individual R,G,B curve manipulations all the time when I am trying to tweak color, particularly skin tone. But, I'm not ready to use individual channel moves for tweaking contrast unless I've got them in luminosity blend mode. I really bought into the concept of separating contrast from color in Dan's Lab book as it really, really seems to simplify things and I'm not aware of any quality or flexibility sacrifice for that simplicity.

Nikolai · December 31, 2006

Andy,
Great job on the chapter!

Speaking on per-channel curves and such...
1) I understand that master RGB/CMYK curve may, and most likely, will shift the colors. Hence - adjustment layer in luminosity mode or immediate Fade-Luminosity.
2) I also understand that a cleaner, yet still rather simple, way to do a basic curve work is doing it in L channel (adjustment layer or not)
3) What I do not get, not at least at my "spine brain" level, is how Dan comes to all those very particular and very different per chanel curves.
I mean, change any one channel - and your colors are shifted all over the board. He speaks of trade off, and that's exactly what we get here: if we improve the color/contrast/perception in one area, we inevitably lose something (shift color, for instance) in some other area, or, most likely ALL other areas of this image..

At this point I'm kinda in boat with Andy: if I need to do something curvy in one particular area and I do not want this change to affect the rest of the image - I simply do it via selection, or, actually, on a separate adjustment layer (which has mask by default) in a luminosity mode.

PS. I'm 40% done with my chapter (6, sharpening), hope to be ready soon.

Andy · December 31, 2006

Granted, this may not be the best example shot (Tailor-man) but I'm not convinced yet by anyone that individual moves are worth it.

edgework · December 31, 2006

Andy wrote:

Granted, this may not be the best example shot (Tailor-man) but I'm not convinced yet by anyone that individual moves are worth it.

If you feel that the image has no cast, then a contrast move is all that is called for. If there is a cast, then the Master curve will simply enhance it.

AZsnapper · December 31, 2006

Curves
Since reading this book, I have been trying real hard to only do per channel adjustments, and use the Shadow/Highlight tool a little bit.

I find the per channel works rather well if I know there are true blacks, whites, and neutrals in the image. But it can sometimes get a little frustrating.

edgework wrote:

If you feel that the image has no cast, then a contrast move is all that is called for. If there is a cast, then the Master curve will simply enhance it.

Andy · December 31, 2006

edgework wrote:

If you feel that the image has no cast, then a contrast move is all that is called for. If there is a cast, then the Master curve will simply enhance it.

Heheh I'm actually tired of him (the suit-seller), I originally processed him in BW and I really like him best that way...

kwalsh · December 31, 2006

I've played with a few more of my images and re-read the contrast chapter. A few things I've noticed:

1. It is very hard for me to do different curves per channel without using a luminosity blend. I'm definately not talented/experienced enough to do color and contrast at the same time.

2. In the end I haven't found an image that I see a drammatic difference between per channel and composite curves. HOWEVER, for me at least this has only been true when I do the per channel first and then play with the composite. I definately find it easier to create maximum contrast on the subject using per channel curves and luminosity blend. After doing this I can usually pull off a composite curve that does nearly the same thing. I have trouble eeking that much contrast out of the subject just starting with composite blind to the real possibilities in the image. That said, I rarely actually want that much contrast and I dial back the layer quite a bit. IN THE END, at the contrast levels I want if I really search for it I can see a very minor benefit in the per channel curves, nothing major.

3. Reading through what Dan says in the chapter I think perhaps some of our perplexity results from us doing RGB curves. It sounds like the master (i.e. composite) curve in CMYK is what really needs to be avoided ("disasterous" in Dan's words) because in that case the black channel really is totally unlike the others and almost always must be handled differently. He states in RGB this isn't as much of an issue except for a less "optimal" handling of contrast and the bigger issue of color and saturation shifts. I'm thinking composite curve with a luminosity blend is probably the fastest solution.

4. Now that I am aware of it I'm quite amazed at just how much the composite curve can freak out color and saturation. It think it definately should be used with a luminosity blend. It also makes me fear whether I'll ever get curves to work on color and contrast simultaneously. Composite seems so benign and yet it can really screw with the image.

5. With regards to the "trade off" Dan does address this in one of the early chapters. He basically says yes, you may freak out other colors with your curves. That is only a problem if those colors actually exist in the image you are working on. He's got an example (I believe it is the purple dress again) with a macbeth color chart in it. In his curve setup a lot of the colors on the chart go wacko. So what, they aren't part of the image. I think his point is if there is one dominant color in the image you are going to need to expand it (sort of like LAB does) to match the eye's simultaneous contrast and in doing so you are going to trade in screwing up a lot of other colors. That's OK if those colors aren't around to begin with.

Ken

Duffy Pratt · December 31, 2006

The by-the-numbers approach doesn't care about color shifts unless it does one of the following:

1) Kills neutrality in obvious neutral areas
2) Throws known colors out of whack; or
3) Makes something in the image have an impossible color.

From this perspective, there is a great deal of leeway for most things in most images. In the tailor shot, take a look at the yellow shirt in the right of the various submissions. No one knows what color yellow that shirt is. We do know it is not the color in the original picture, because of the cast in the entire picture. So how much of a move makes a color shift to that shirt untenable? This is a question on which one can disagree.

The samoyed shot, for the same reason, is much harder to color correct. Snow is white, and so is the dog. but if you look at the original, the snow is anywhere from -4 to -8 in the B channel, while the dog is anywhere from -3 to 4. That's a fairly big spread of 13 points in the B channel for stuff that is all supposed to be neutral. To make matters worse, it doesn't appear to me that the variation can be explained by light and shadow. On this picture, I just decided that people can live with blue snow, since all sorts of shots have that as a convention, and because the dog is the subject of the picture. But blue fur is a big no-no. So I killed the blue in the fur and generally cooled the picture. With more time, I might try to kill the cast in the snow and the fur, neutralizing them both, while retaining the brownish red in the dog's eyes and on his nose. That takes techniques that are outside the scope of this chapter, but its not that hard to do.

Duffy

rutt · January 2, 2007

OK, Andy, you question two of Dan's points:

Prefer global moves (curves, blending, sharpening, that sort of thing) over local ones (selections, painting, burning, that sort of thing.)
Superior results are possible with by writing a separate curve for each channel than by using the master RGB curve.

These are both excellent questions. I think the answer to 1. should wait until we've absorbed more of the book. When only a local move works, Dan doesn't hesitate to make it, but he accomplishes a surprising amount without doing so and mostly it is easier (once you know how!) and looks better.

The second question requires some careful parsing. If you know how to write individual curves, you can achieve superior results for most images. Sometimes the difference is very subtle, especially when a single color is most important, as in portraits and especially on a monitor as opposed to in print. How much better is Dan's waterfall with individiual curves than with the master curve? A little. How much better is Dan's image of the mountain in the mist with the individual curves than the version with master curves. A lot, I think. At least a lot by Dan's standards (remember, this is a competitive sport for him.)

I found an example where I could do a lot more with individual RGB curves than with the master curve. Here is the original:

With a master curve adjustment to set a light point in the stone just in front of the chestnut pan and steepen through the subject (face, shirt, fruit, chestnuts, apron; this is a busy image:

With individual curves, I was able to steepen each element of the subject more carefully without blowing or plugging other parts:

Now the chestnuts, fruit, shirt, apron, and face all pop better. [No plate blending, here. No sharpening. No LAB or CMYK. No exposure compensation. I'm just trying to compare apples to apples re master vs individual RGB curves.

If you only use the master curve you are limited to fairly simple forms of horsetrading. Often that's fine. But learning to use the individual curves can pay off by letting you squeeze a little extra contrast here without having to pay for it there.

Others have already pointed out that using individual curves works better when you are fighting a cast or want to change the color balance in some way. We can all agree on that, but that's not what this chapter is about. Dan's point here is that you may be able put together a complex trade that doesn't really lose anywhere in order to pay for it's improvements. But, the best examples of this are going to be "busy" images.

It was interesting to read this chapter again today. RGB curves are an elementary tool in Dan's box, but this particular application isn't my most common use for them. These days I do a lot of channel blending to try to bring out contrast and use RGB mosty in color mode to correct color balance. But Dan's images of the waterfall and especially the Chinese landscape show just how far this technique can take you if you are willing to think about the channel structure before starting to write curves.

Chapter 2, Professional Photoshop, 5th Edition

Comments