Been awhile, hasn't it? Yeah, I've been pretty quiet this past while. Oh, the places I've been and the things I've seen. Quite a few things on my mind. Let's see if I can take you down one of my roads.
Personally, I don't put a whole lot of stock into Histogram. It has it's uses, but I don't swear by it. But frequency has been on my mind an awful lot and it has led me back to Histogram. In particular, Equalize. There is a connection, and it has to do with what I call the human touch.
Since I want to talk about masking, how about a quick review?
A mask a greyscale. That is what we want to end up with. The question is: what data set do we use to get greyscale? The answer, usually, is Luminosity. Why? Because of our eyeballs and because it usually the better data set.
How do we get Lum?
- copy photograph
- Edit > Fill
-- Using: Black, White, or 50% Grey
-- Mode: Saturation
Tada. Instant Lum channel. But you knew that already.
From here, we want to divide the Lum into highs and lows. Most people would just use Levels or Curves. Maybe even Threshhold followed by Gauss. But we don't want to do that just yet. Why? Because we are after the frequency of the entire photograph as a whole. Ah, that is a magical quandry. How do we get the general frequency of an entire photograph without the pit falls of frequency? I'm sure you are aware of the infamous halo quandry when using Gauss et al for masking. Maybe we can take care of both quandries in one fell swoop, eh?
Sure can, and all it takes is Image > Adjust > Equalize.
What you will be left with is the statical top half and statistical bottom half.
- extract Lum
- Equalize it
- use it as a mask for your favorite techniques
-- load it straight-up for highs
-- invert to get lows
I have used this on many washed-out photographs with *very* pleasing results. Even used simple variations to take care of stark and hard contrast. Let's just say that I'm very happy, and that doesn't happen very often.
Being a PS neophyte, I was mostly lost until this part:
quote: warjournal said:
I have used this on many washed-out photographs with *very* pleasing results. Even used simple variations to take care of stark and hard contrast. Let's just say that I'm very happy, and that doesn't happen very often.More to come.
Set as a layer with blend mode of luminosity did a nice job for a wishy washy photo of mine. (Opacity something like 60-70%)
Overlay was too strong.
Pretty nice wj!
Okay, time for a little detour. This is one of my favorite techniques, but I don't think I've really talked about it.
Now, I'm a big fan of using Lum to generate masks. It's greyscale and the data is usually pretty good. Sometimes I'll manipulate RGB before extracting Lum, but mostly use Lum straight-up.
Once you get Lum, how you use it to generate a mask is a matter of taste. Maybe Curves, Levels, and toss in some Gauss. Or maybe even use Find Edges and/or High Pass to generate a surface mask.
For me, one of the ways I choose how to tweak is by examing what I'm after. Well, duh. What I mean is that you can divide a photograph up into highs and lows. You know, two major chunks. Or you can divide a photograph up into highs, midtones, and lows (HML). Now, dividing up into 3 chunks can be tricky.
For getting HML, I absolutely detest Colour Range and using the drop-down. I detest Colour Range for a lot of reasons. I do use CR on occassion, but *never* for HML.
You can use Curves rather easily to get HL in HML, but using Curves to get M in HML isn't really that great. It can be done, but the B-Spline can cause serious problems. This is especially true when you want to custom define the midtone. If you try to use Curves to make the midtone 180, the B-Spline will kick in and be not so good. Ugh.
What is the WJ method for getting HML? Gradient Map. Quick-n-sleazy, lemon squeazy. All you need is 2 or 3 stoppers.
Highs:
- RGB (0,0,0) @ Loc 50%
- RGB(255,255,255) @ Loc 100%
Mids:
- RGB(0,0,0) @ Loc 0%
- RGB(255,255,255) @ Loc 50%
- RGB(0,0,0) @ Loc 100%
Lows:
- RGB(255,255,255) @ Loc 0%
- RGB(0,0,0) @ Loc 50%
I've got a crap ton of G-Map presets of those and variations. Uber easy with nice and tight linear blending. Oh, speaking of linear blending, I always make sure that Smoothness is set to 0%. I *hate* B-Spline causing overlap and messing things up. Major ugh!
Yeah, I use Gradient Map quite a bit. I've got tons of presets saved which are right there in the dialog, and giving the stoppers some wiggle is no big deal. And there is the added bonus that G-Map already works on Lum.
When I get the yarbles for it, I'll toss up some images to help illustrate this little bit.
At the risk of sounding really stupid, how do I go about generating a mask from the Luminosity I obtained by following your instructions in the first post? I thought I knew the basics of masks, but I guess there are some major holes in my knowledge.
If the answer to this is going to be part of the "tossing up images to help illustrate this" then I'll wait.
Or if someone wants to give me a quick pointer in the meantime, I would be obliged. Sorry for being so clueless. Everyday I found out new stuff I didn't know about PS.
Suho, it's all about Channels palette and selections. I dare say it's even about Layer Masks, which are a part of Channels.
Try this:
- open some random photograph
- copy to new layer and use Edit > Fill to get Lum
- in Channels palette, copy R, G, or B channel of Lum layer to new alpha
- turn off Lum layer in Layer palette
- in Channels palette, go to the alpha channel that you copied
- Select > Colour and use drop-down to select Midtones (you know, marching ants)
- go back to the Layer palette
- add a Hue/Sat adjustment layer and raise saturation to taste
What did we just do? We used Lum (rather directly) to select midtones, and then used that as a mask to raise saturation.
warjournal: Yes, that makes sense. Thanks. I was just wary of using Color Range because I had just read above that you detested using it for getting HML masks, and that was the only way I knew how to get a mask.
Follow up question: I tried a midtone mask on the Lum alpha channel and then another on the original photograph (using Color Range). Is there a difference in what I will get? Just from what I can see, the mask from the Lum alpha channel and the mask from the original appear to be pretty much the same.
I'll just hang around and try to follow as things develop here. A lot of what you talk about is way over my head, but I'm always eager to learn new tricks and techniques. When Steve introduced me to the wonders of Curves it pretty much changed my life.
quote:Follow up question: I tried a midtone mask on the Lum alpha channel and then another on the original photograph (using Color Range). Is there a difference in what I will get? Just from what I can see, the mask from the Lum alpha channel and the mask from the original appear to be pretty much the same.
A Gradient Map works on Lum already. Extracting Lum and then using a G-Map is rather redundant, but I usually do it anyways. If you use a G-Map that goes black <> white and Smoothness set to 0%, then you will get Lum. This means that you can keep a G-Map in the Layer palette. Which means you can turn it on/off and do various other Lum mask tricks on-the-fly if you want. Not just turn it on/off as you see fit, but also change presets rather quickly. Very handy.
Something like:
RGB 3d > Lum 1d > RGB 1d/3d
This is why you can't use Image > Adjust > Gradient Map in the Channels palette - it needs to start with RGB to get Lum and interpolate the Lum line to get final RGB. You can't use RGB in the Channel palette, either going in or coming out.
(If you get into skeletalization or Cellular-ness of 3d space, you can create a true 3d gradient map. Not that I think about these things.)
One thing that I have never heard mentioned is order-of-operattions when it comes to masking. Being aware of order-of-operations can make a tremendous difference. This can easily lead to a phenomenom that I call Doubling Back. You can do this easily with Curves by getting a little bit crazy with the curve. You can even do this with Displace.
In Curves, imagine you have anchors that are at 64 and 128. To see the Double Back phenom, just move the 64 anchor above 128. It will cause, I don't know... solarization? I call it Double Back. This can also be seen in Displace, but I'm not in the mood to show that.
Another place where this occurs is when masking adjustments. When you mask an adjustment, you are moving certain values, and it is easy to move those values beyond values that you are not moving in the mask.
- start with a photograph with decent contrast
- Colour Range > Midtones (even though I hate Colour Range for this, it illustrates very nicely)
- use that as a mask for Levels Adjustment Layer
How far can you move the Gamma (midtone) slider before you get solarization or Double Back?
This means that you have to get the masks and the adjustments to play nice. Getting the adjustments to play nice isn't too hard, but getting the masks to play nice can be a bit of a pickle. To get the masks to play nice, I use the 100% rule.
Let's switch to a bit of a sloppy Curves view. Horizontal is Lum and the veritical is mask values.
Let's say that you want to divide a photograph into highs and lows in a straight-up manner. That is, no major tweaking. Easy enough, and it looks something like this:
Notice that high + low = 255/100%. This is good, and the masks will play nice.
Now let's divide a photograph up into high, mid, and low. If we use the above high/low and just toss in mids, we get something like this:
As you can see, high + mid + low != 255/100% at all points along the Lum line. In this manner, it can be *very* easy to overly double-adjust values and cause solarization.
In order to divide a photograph into high, mid, and low with masks that play nice, you have to use something more like:
If you go back and look at the Gradient Map settings that I posted earlier, you should see that they are consistant with the last graphic. That is, the highs and lows are cut-off at the 50% mark.
Order of operations, eh? Yes, can make a big difference.
Let's say that you divide a photograph into highs and lows and use those as masks. This is largely what I've been talking about. Let's say we change the order a little bit.
- extract Lum
- generate a mask for highs
- using the tweaked photograph, extract Lum again and mask lows
Because some of the values got tweaked after tweaking highs, the lows will be changed a tad. This means that the mask for lows can very well be very different.
Largely I consider this bad. But not always. Consider Shadow/Highlight for a moment. It will divide a photograph into highs and lows, then - and only then - get the midtones.
Get HML in one swoop, or adjust HL and then get midtones... ? Big difference.
The paradigms of masking and order of operations holds true for many things in Photoshop. What happens if you use Levels and Hue/Sat in different order? What happens if you use Levels and Hue/Sat in different order with masks? Hmm... food for thought.
This whole thing is one of the reasons whey I adore using Equalize for masking purposes.
However, the whole thing does fall apart with the addition of one simple concept. And yet it still kind of holds true. Pretty funny, actually.
One thing I am always checking is order of operations with various things.
- extract Lum
- Equalize
or...
- Equalize
- extract Lum
I'm starting to like the latter more than the former.
Also, there is a difference between Equalize on RGB, and Equalize on seperate R, G, and B channels. Seperate R, G, and B can give a good running start on colour balance tricks and ChOps.
I'm a bit lost right now. Too many directions to go in, and some of them are only mildly connected.
I think next I'll toss out my notes on Smoothifying. Just to get this out of the way.
Then another Linear Light trick that I totally adore.
Then more notes on frequency and contrast. Surface masks and masking.
By the time it comes to tie in some Equalize, hopefully it will be too redundant.
Many moons ago, I was into deconstructing HSL and simliar spaces. Hell, I was into deconstructing several colour spaces to sort out anomolies. Eventually it led to me to various techniques for extracting raw hue and saturation. Both of these channels were usually very blocky, which is consistant with compressing colour information more than lightness information. Once again, our eyeballs.
I believe it was Deke that popularized an aspect of this. If you super saturate a photograph with even mild compression, what happens? The compression blocks become very apparent. I believe Deke's method uses Median and Gauss to fix.
But what led me in a circle about Deke's method was his use of the Colour blending mode. Since I had become adept at extracting Colour = Hue + Sat, I knew exactly what was going on. He was using Median + Gauss to smooth colour information, and then putting it back in before uber saturating. Right? Right.
Good and all, but I see several potential problems. The major problem being one of lines. As my old art teacher used to say, "There are no lines in real life - only changes." (Interestingly enough, that is why he didn't consider caricature a form of art.) While Median does (largely) respect lines, Gaussian Blur doesn't. Few other things that I don't like, but I'll keep them to myself.
So we have all that blocky information that we want to be smooth. But we want to control the smoothness. You know, smooth in the flatter areas, but we don't want areas of contrast to bleed together. Quite often not necessarily true for colour work such as saturation, but we are looking to expand on an idea. Need a bit of room to flex and adapt. Smoothifying is a grand idea that is just begging to be expounded upon.
Enter, of course, Lens Blur. With Lens Blur, we can put the lines wherever we want. We can make areas play nice and not play nice as we see fit.
How do we go about making those lines? Of course you can draw them by hand as you see fit. Or maybe there is a more automated way. In an automated way of finding lines, there are more questions. Ugh! What data do we use and how do we manipulate it to get the lines we want?
Now, there are two main branches of this technique that I use. I'm only going to talk about one of them.
We are looking to generate a surface mask for use with Lens Blur. The mask in it's final form has to be greyscale. That right there answers the question of what data to use: Lum, of course. It just plain has the best resolution, so to speak. Use the better data to make the not so good data a little better.
How do we manipulate Lum to get edges? For that, I'll point you to Surface Mask. In a nutshell, use Sobel and LaPlacian to compliment each other. They do a very fine job when working together.
Of course, I do have a beef with the LaPlacian method. That being that not all frequencies are equal across a photograph. On occassion, when I feel the need, I'll use Lens Blur to get variable High Pass, and then use that to get variable LaPlacian edges. Not often, but the human touch is required sometimes.
What can you do with a surface mask? All sorts of things. For example, Equalize can seriously bring out compression badness, but a decently crafted surface mask can help ease the pain. Can also help with uber saturation (Deke). And a few other things, but I'll leave it to you to find them and explore.
quote:Now, there are two main branches of this technique that I use. I'm only going to talk about one of them.
Just for shits, gonna talk about the other main branch that I use.
One thing I absolutely adore is extracting Lum and then using that. It has its advantages. Unfortunately, it has its disavantages. The big disadvantage being that it's 1d. On the other hand, RGB is 3d. We can take advantage of the 3dness of RGB when it comes to some things.
- start with some random photograph
- copy to new layer
- Filter > Blur > Blur just get rid of some noise
- Filter > Stylize > Find Edges
What are you now looking at? You should be looking a bunch of white, some black lines, and some lines of various hues. What if you want a yellow line to be just as black as a blue line?
Ah, the subject of desaturating with intent has just popped up. I'm trying to keep my feelings out of this, but a great deal of people out there need to STFU when it comes to desaturating. They have no real grasp on contrast, data, or tools. Fuck 'em right in the ear.
Let's say you have a photograph of a flower. The petals are blue and the background foliage is dark green. Two seperate elements, pedals and foliage. If you desaturate something like that to Lum, chances are good that the Lum will be a bit washed out between the two and not the best for finding edges between them.
Two obviously different hues, and we can use RGB in 3d to find the edges between them. All you have to know is your ChOps. Remember Finding Min and Finding Max? Disco if you do. That will take care of the hue and edge problem. All of the edges will have maximum difference from pure white.
See? Decide what contrast you are after, see how it is manifested in the data, and then use to tools to amplify it. Rant! Cuss! Rant! Cuss! Stomp feet! Scream!
That was Sobel, or Find Edges. What about High Pass using the same idea of maximizing difference?
- random photograph
- copy to new layer
- Filter > Other > High Pass to taste
- Edit > Fill
-- Use: 50% Grey
-- Mode: Difference
- ChOps to Find Max
Ta-freaking-da. Now ready for further tweaking as you see fit. Levels, Invert, whatever.
We took advantage of RGB and found maximum difference in Find Edges and High Pass. This is good when you have lots of hue. Also good with diagrams, logo-ish things, and signs.
There is a great deal more to contrast than a handful of the same ol' tricks. And, of course, if you understand what I'm babbling about, you should be able to adapt things to your own flow as needed. You don't always need maximum difference in Find Edges, and you don't have to use Finding Min/Max to get it.
How about a tangent? Or a detour? More junk? I know you want more.
Have you ever looked at a technique and thought to yourself, You know, I could use that for something entirely not related to the turorial? By God, I hope you have.
I adore Gauss. Raise your hand if you adore Gauss, too. Raise your hand again if you get tired of frequency not being equal and causing halo problems. Hmmm, tough one. Try this: raise your hand if you hate halo. Better yet: raise your hand if you adore Gauss and hate halo. Ah, thought so.
Have you ever looked at a technique and thought to yourself, This might help me get rid of halo?
Hmm... halo and totally unrelated techniques - let's get married!
Raise your hand if you've seen softening techniques involving Gauss and Overlay, Soft Light, or whatever.
- copy photograph to new layer
- Gauss to taste
- change blending mode to whatever and reduce opacity to taste
Sure. Wonderful. Raise your hand if you've used this technique before. Now, raise your hand if you've used this as a masking method to help get rid of halo. Ah, the light is too bright and I can't see your hands.
- copy photograph to new layer
- extract Lum using Fill
- copy Lum to another layer and turn off top Lum layer
- back to bottom Lum layer and Level the shit out of it
- give the Leveled Lum some Gauss
- back to top Lum layer
- mess with different blending modes and opacity to taste
Eventually you should end up with some softness and some detail. With a little practice, can be a damn fine mask with little or no halo for those funky tweaks that you love and adore so much.
That's about it in a nutshell. Believe me, I could write tons on just this little bit. It gets really interesting when you start using it with intent.
From: The Land of one Headlight on. Insane since: May 2001
posted 10-04-2006 08:55
And we bow in your general direction
___________________________________________________________________________
The goal in Life's Journey is not to arrive at the grave safely in a well preserved body, but rather to skid in sideways, totally worn out, shouting "holy moly what a ride!"
I want to talk about frequency a little bit. Not frequency in the way that you have heard me talk about frequency, but in a delta sort of way.
I draw things in triangles, circles, waves, and similiar. If you give me Levels, I'll draw some triangles and trapezoids to explain it. Triangles is how I see ratios in my head. Give me Hue/Sat, and I'll draw an offset wave and some circles to explain how I see it in my head.
Offset wave? WTF is that? It is simply modifying data, usually with a mask. That's it.
Consider the gamma slider in Levels. If you move it one way, then you are adding to the midtones, but adding nothing to black or white values. If you move it the other way, then you are subtracting from the midtones, but subtracting nothing from the black and white values. Even though not technically correct, gamma can be thought of as an isosceles triangle. This is how I see it. More in the middle, none at all at the ends.
One day I was working on increasing tonal range in a photograph. Sometimes subtle shades of orange just aren't enough and I want to punch things more into the red and yellow range. So there I was with my low-tech notebook and pencil just drawing away when it occurred to me. I can trick Photoshop into working the exact same way that I think.
This is the way that I think and see things. Gonna switch it up that basterd Curves view that adore. Now, the numbers on the side are for illustration. I'll get to those in a minute.
Upper-Leftt
This is how I see Brighness. No matter what value you put into it, it will add approximatley 64. This is how Brightness in Brightness/Contrast works.
Upper-Right:
This is basic contrast like the Contrast slider in Brightness/Contrast. It will add nothing to values in the middle. It will add as you get higher, and add even more as you get closer to 255. Same thing with <128 end -- the lower you get, the more it subtracts. Basic scaling, and a great way to blow your highs and lows.
Lower-Left:
This will decrease contrast much in the same way as adding contrast. Nothing in the middle, subtract more and more as you get higher, and add more and more as you get lower. This will scale things tighter to 128. This is a great way to wash out a photograph. Interestingly enough, you can use this to invert the data.
Lower-Right:
Ah, a softer offset wave for contrast. This will add to values that are around 192, and will subtract from values around 64. This will do nothing to the highs, mids, or lows that are input into the wave. So, this is a way of adding contrast without blowing the highs and lows.
The last one, the lower-right one, might look a little familiar to some of you folks. Either the wave shown, or the description. Now, I called those diagrams a basterd Curves view for a reason.
Aha!
Did it click? Did you realize that you were looking at a rotated view of Curves? You get +5 WJ Points if you did.
How do we get Curves to behave like in the basterd, rotated waves? Quite simply, Linear Light. Start a Curves Adjustment Layer, flatten the curve to 128 all the way across, and change the blending mode to Linear Light.
But LL is a bit harsh because of the doubling. In order to halve the doubling, simply bring the Opacity down to 50%. With this last little bit, you can now work with delta directly. And, for a guy like me, in a way that is 100% consistant with the way I think on a fundamental level. Well, not that I can't think in Curves as-is, but sometimes Curves in LL helps me depending on my mood.
If you understand using Curves in LL, then you understand calculus. If you think you are a numbnuts that can't understand calculus, maybe you aren't as dumb as you think. Or maybe you just need a different tutor. Heh, using Photoshop to teach 'advanced' math concepts. I find humour in that.
Of course, the usual what works for me may not work for you. Take the bits and refine your own flow. Find your center and unleash it.
There is another reason why I used those diagrams. This will be next and it just might prove interesting.
One thing I tend to do is run in circles. Sometimes I really hate this. I have a quandry, start exploring, and eventually end up right back where I started. Answers lead to more questions. Several answers down the road, I'm back to the original question. Ugh.
But sometimes I really enjoy running in circles. This is like a satisfying game of Connect the Dots. Rather, Connect the Concepts.
In that tutorial there are three graphs. The first graph is a gradient. The second graph is the first derivative. As luck would have it, the original gradient gives the second graphic (first derivative) the appearance of the infamous Gauss bell curve. Coincidence? Hmm...
Now, the third graph is the second derivative. Does it look familiar? Does it? I showed a sloppy offset wave that looks just like it. Not only that, but it happens to be High Pass. Did you know that High Pass does the second derivative of a gradient using the Gauss bell curve? Of course you did.
And all that crap comes around full circle. I'm very satisfied with this round of Connect the Concepts. Bloody awesome.
Now we are going to pull the contrast aspect out and play with that for a bit.
Contrast, to set in opposition in order to show or emphasize differences. That's it. Pretty simple, really. Honest. I wouldn't lie to you. At least, not intentionally. But even then I'm prone to addendum to patch things up. Right?
The sloppy contrast wave that I showed was for an image on the whole. Easily done with Curves. Or you can use Levels or Brightness/Contast for sloppy contrast. This is a popular way of stretching data to emphasize difference.
The second derivative showed in the tutorial is High Pass. This will increase contrast using a specific frequency based on surrounding values of a given pixel. Ah, yeah. Even though this is a form on contrast, it is commonly known as sharpening.
Ah, frequency and contrast. Lovely subject.
A lot of people using various frequencies for various sharpening tasks. Consider sharpening eyes but not sharpening the acne on the forehead. Sharpen, then use a mask to keep the sharpening that you like. Masking a sharpening frequency is well known in most retouching circles. But there is still so much more to be explored and played with. Masking a frequency is only half of the story.
I'm going to repeat that. Masking a frequency is only half of the story.
Have you ever considered manipulating the frequency itself before using it to sharpen? This is the other half of the story.
Even though clipping usually isn't a problem when it comes to sharpening, I'm going to use clipping as an excuse to show the basic technique.
Where in an image can you add and subtract to cull detail without clipping the half of the detail in the highs and lows? In the middle, of course. (If you are working on midtones and clip under or above, you might want to re-examine your flow. And I thought I was crazy. Egads!)
- some random photograph
- favorite method to select midtones
- use said selection to mask sharpening
Using that, you should be able to keep everything legal, within bounds, or whatever. Now, this might not exactly make sense in a practical manner, but beer with me for a few.
Why mask out the HL and keep only to M?
Let's say you have a photograph of some sheer curtains over a bright window. Rather high lum with some subtle detail. You want to pull out some of the detail, but you don't want to blow out the highs because they are already pretty close to being blown out. If you sharpen the curtains, you are going to clip some noise right over 255 (or whatever number you use). In a case like this, prolly Levels, but might not be enough as you want to punch some detail as well stretch the curtains as a whole.
Time to modify the frequency itself.
Well, crap. Even though I just talked about using highs as an example, the graphic that I prepared uses lows. Heh, that's what I get for listening to Barry Manilow while flowing. Take everything I just said and flip it.
So, our basic technique evolves a little bit by simply modifying the frequency to take the subtract parts out. Taking the subtract parts out for shadows, and my layer rig looks something like this:
I prefer Linear Light and flattening to 128. Why? Because wiggling the anchor that's at 128 can be fun. But feel free to mix it up.
Blondie just started singing. I think I need more Korn in my diet. Meh, I'll switch to Queen for the next few minutes. There goes my baby. She knows how to rock-n-roll. She drives me crazy... Ah, much better.
What was I talking about? Well, shit.
Levels and a modified High Pass wave. I have used this combo many times to get far more detail out of the shadows than either would allow solo. Can pull things out a great deal before the data stretching starts getting nasty.
And he nails the dismount! The crowd goes crazy!
Now get to playing before I get my chainsaw out and start breaking stuff. Ah, that's the Limp Bizkit talking. I think it might be time for some Tequila to mellow down.
OMG I'm a nut. Next time I flow, take my WinAmp away.
Curves and LL:
Try Curves + LL in Lab mode. In LL or not, if you are using Curves in Lab for colour, try reducing the opacity down if you are not in this particular habit. Using Curves + LL in Lab for colour, I usually go around 25%.
In RGB, you can get use Curves + LL to work on Lum.
- start up your Curves + LL
- new layer and fill with any shade of grey of you want
- set blending mode of this layer to Saturation
- clip it to the Curves LL layer
I think I had a few more random notes and parlour tricks, but it's time to be Down with OPP.
In case you couldn't tell, I like Linear Light. Why? Because it's uber easy to understand. It starts at 128 and the fall-off is linear. No funky curves or conditionals involving the substrate. Easy.
I will be talking about saturation. The kind of saturation that I will be talking about is the kind that is used by Photoshop in HsL.
Now, what I want to talk about next is something that *seriously* gets under my skin. This is something that gets under my skin far more than quarter-brained tutorials on Displace. Good chance I'll be coming off a bit angry. Even if I let this cool, revisions would still be angry.
Let's talk about contrast and saturation.
One day I was cruising around reading random tutorials about sharpening. Came across a tutorial that went something like this:
- copy photograph to new layer
- High Pass to taste
- set blending mode to Linear Light
The author then went on to say something about how LL will over-saturate with HP like this. His answer? Change the blending mode to Hard Light.
Ugh! Jackass!
But it was just one tutorial and I didn't fret over it. Then I read another tutorial that went something like this:
- Unsharp Mask to taste
- Edit > Fade to Hard Light (or something like that)
Ugh! Again!
You might be considered a guru by others, but don't you dare call yourself that around me. I hate memes that suck beep. Okay, WJ, take a deep breath. In... out...
What is contrast? Finding difference and exemplifying it.
What is saturation? Formulatically, something like this:
Sat = Max(RGB) - Min(RGB)
Sat is just simple difference. The higher the saturation, the greater the difference. The lower the saturation, the lower the difference.
Saturation is based on the seperate R, G, and B channels. If you increase saturation, then you are increasing the contrast between the individual R, G, and B channels.
What does this mean? This means if you manipulate R, G, and B channels seperately, then there is a good chance that you will lower Min(RGB), raise Max(RGB), and thusly increase saturation. When you use High Pass in RGB, then you will raise saturation where it already exists.
This is also why Curves, Levels, and Brightness/Contrast can increase saturation so dramatically as well.
The answer? Add or subtract the exact same value from R, G, and B. In this way, the difference, or saturation, will stay the same.
How do you do that? The easiest way is to extract Lum, sharpen that, and then put it back in (Lum blending mode).
* I also hate people that say that you can't sharpen RGB without over-saturating like you can sharpen Lightness in Lab without over-saturating. Whatever. "Oh, I sharpen in Lab mode to avoid to over-saturating. And then I convert back to RGB mode." Guess what, numbnuts? You can do the exact same thing in RGB without over-saturating. You are wrong with a capital F. Breath in... breath out...
Another way to sharpen without over-saturating is with Linear Light.
- copy image
- Edit > Fill to extract Lum
- High Pass the Lum
- set HP Lum to LL and reduce opacity to taste
Since you are running HP on greyscale, the resulting frequency will be greyscale. This means that the exact same value will be added to each RGB channel. Saturation stays the same.
Ta-beeping-da.
Okay, I'm not telling the whole truth.
1. If you use Lum to sharpen, you might reduce saturation in the highs and lows. Largely unnoticable.
2. If you use HP to sharpen on greyscale, you might shift hue a little bit in the highs and lows. This might cause noticable shift if you are working with a lot of bright hues. This goes for some of the other blending modes as well.
That's good and all, but the name of the game is contrast. Saturation is a form of contrast, and we can take advantage of that. If you are trying to pull more detail out of shadows, then let RGB > HP do it's thing and increase saturation. Why stretch data in only one channel when you can have all three? Increase contrast like this, increase contrast like that, and let it all build up. You just might be amazed at how quickly different kinds of contrast add up, even if you take it easy on the different kinds of contrast. Just know the data and how to pull it around.
Try this. In the midtones, sharpen using Lum. In the highs and lows, sharpen using RGB.
I'm so full of crap right now. Lots of small things on my mind and a few big things. I want to get some of the smaller things out.
The tools in Photoshop have rather specific uses. Some of them are obvious and some of them are not so obvious. And then you have to deal with how the tools actually work.
For example, you can use Levels to add contrast to a photograph. If your highs are a bit low, off to Levels to stretch that data to brighten it up and add some contrast. Another use for Levels is colour balance. Easily done with the gamma sliders in Levels.
One of the things about using Levels for colour balance is the axis upon which you are shifting the colours. If you wiggle the Red gamma slider, you are adjusting colour balance between red and cyan. If you wiggle one gamma slider, you will move colour balance between R/C, Y/B, or G/M.
What if you want to adjust colour balance with Levels, but you want to betweeen RY and CM? Well, you would have to wiggle more than one gamma slider and get them to coordinate to get the variable axis that you want. Unless you are Uber, and I'm not when it comes to this, can very well lead to a lot of back-and-forth between channels in Levels.
Bah! Just shift the data, tweak, and unshift the data.
- Hue/Sat Adjustment Layer using Hue -30
- one top of that, another H/S using Hue +30
- toss Levels Ad-Layer between, and just use the R gamma slider
With that, you can adjust the colour balance between RY and CM with just one gamma slider. Simple, yet snazzy. Coo coo ca choo.
Another place where I've used this basic idea is with exploding data. Explode the data, tweak, and unexplode the data. Shift it big, tweak, then unshift it back down to small.
Let's say I want to use Curves to tweak the range 100-150. Well, that's not much room to move in Curves dialog for that short range. I'll grab the range, explode it to 0-255, tweak away using full Curves dialog, and then bring it back down to 100-150. It's a bit much, but sometimes uber fine-tuning the subtleties can be good. Not that I'm that kind of freak, mind you. Okay, maybe on the rare occassion I am, but generally I'm not.
Just a simple parlour trick that may come in handy. But it's up to you to spot when to use it. I only gave two quick examples, but there are plenty more possible applications out there.
On with another parlour trick.
One thing I absolutely abhore is the recommendation to use Channel Mixer to greyscale RGB. Bah! Leave greyscaling to better methods and leave CM to mixing channels.
One fine day I was messing with making things fractal. Things like Mosaic and Cellular. I kept thinking about mixing the various levels of data when it comes to fractal things. Data? Isn't a channel nothing more than data? Why not toss the data that I want to mix into Channel Mixer and let that do the mixing for me?
Now there is a grand thought for ya. Channel Mixer doesn't have to be limited to RGB of a photograph. After all, you can cut-n-paste any channel you want into the, uh, channels. Does the R channel have to be the red of an image? Certaintly not. It could very well be Lum. Does G have to be green? No. It could be Lightness. Could be Average(RGB), Max(RGB), Min(RGB), Gauss, High Pass, and so on.
Not a completely infallable idea, but a fun one to play with.
There is a great deal more to Channel Mixer. Not just the good, but the bad. Maybe one of these days I'll tell y'all about it.
From: buttcrack of the midwest Insane since: Oct 2000
posted 10-09-2006 20:05
Ok, I decided to start saving some of this magic warjournal shit, in the hope that someday I may actually understand it all. However when I did a search for "warjournal" it only produced 5 entries. I know there have been more rants than that. WTF ?
WJ : Not sure where all this comes from (besides your undying curiosity and dedication) but I`m glad it comes.
Thank you until you`re better paid.
I'm not sure about here on the Asylum, but I do have a great deal of my stuff archived. Goes back like to almost when I came here. Been here 6 years, so my personal archives go back around 5 or so. I really got to get to getting my junk on wiki. Not just rambles, but more plug-ins, parlour tricks, and interesting observations. Oh my dear, sweat wiki... why have I forsaken thee?
How about a quick chat about Equalize? Sounds good to me.
I'm sure most of you folks are familiar with Auto Levels. Sorely loved, yet delightfully dreaded. Auto Levels will find a high point and a low point using histogram, and then stretch the data based on those points. When you set the clipping percent in Auto Levels, you are dipping into the histogram. In recent versions, Auto Levels comes with various options for what histogram data the ladle dips into.
To see how Auto Levels stretches the data, just use Histogram. Maybe even try different methods of histogram dipping. It should be fairly easy to see how linear the stretching is.
What about Equalize? It stretches the data as well, but in what manner? Again, do before-and-after of Histogram to see.
In using the Histogram palette, you may have to watch the cache level. Depending on cache level, you can get very different histograms for the exact same image. Can't say I know the rules for when the Histogram palette decides to up the cache level. Just keep an eye on the upper-right for the little warning triangle.
Eventually, we get something like this:
In the Manual Levels example, pretty easy to see how the gaps are mostly evenly distributed (I suspect B-Spline distribution of gaps). But how are the gaps distributed in the Equalize example? The mountain of values got stretched, but how did Equalize know?
If you can believe it, Equalize is about probability and manipulating it. The histogram is built, sorted, and then used as a look-up table. That's it in a nutshell. (I had a link explaining this rather well, but it appears to have gone bye bye.) The result is that you should end up with pretty evenly distributed values.
Unfortunately, this method is more of an approximation. In order for this to work with absolute precision, you have to start with an image that is absolutely perfect. Well, perfect in the data, and I have *never* seen that happen and I don't think I ever will. What is the probability of finding an image with perfect data? Heh.
We end up with an approximation and gaps where they 'should' be.
Is there a use for this fugly operation known as Equalize? There is at least one very functional technique that I've discovered, and it gets into midtones.
Snap! Did I just say the M word? I think I did.
If you use Equalize to extract HML, you are doing so based on probability and end up working with percents based on the image as a whole. If you use Equalize to get M in HML, you should end up with approximately 50% of an image. If you get H in HML, you should end up with approximately 25% of an image. And L being approx. the last 25%.
This can be an excessively useful parlour trick. When my eyeballs can't handle getting the midtones that I want, I'll use Equalize to do it's best for me. Mostly in images that are too washed out, or when the contrast is way too high.
Are there any drawbacks of using Equalize? Of course, but I'll leave this to you. Consider this a part of knowing when to use Equalize and when not to use Equalize.
I think I'm done. I had more to talk about, but the urge is gone. Right now I'm getting sucked into more into contrast and cheap ways of faking it. As much as I adore gamma, I don't like all those power functions slowing me down. Back to linear algebra and geometry for some quick-n-sleazy fakery.
Have you ever looked into gamma and human perception? Good stuff. Several places for eyeballs, and even some stuff for our ears.
posted 09-25-2006 00:34
