B&W conversion workflow

rutt

I've been doing a lot of B&W conversions recently and I get a lot of compliments on them. I have developed a workflow that I always use. I think it's really very simple. The results are particularly good for pictures of people against various backgrounds. Once mastered, it takes about 5 minutes per image or less, depending on how fussy you are. Unlike plugins and separate programs, you do have to understand some photoshop commands and color theory, but at least for me, that's part of the fun. As usual, all the basic ideas are due to Dan Margulis.

The workflow has three steps:

1. Convert to B&W by channel blending.
2. Curve to enhance contrast
3. Sharpen two ways for local contrast and sharpness

I'll go through these steps in detail with an example image taken by Alex Feldstein from this thread.:



Unfortunately, it's not full sized, so our results won't be quite what they would be if it were. I'll make an effort to demonstrate on more images in the future, undoubtedly some full sized.

1. Convert to B&W by channel blending
Let's look at the RGB channels:


[size=+2]Red[/size]


[size=+2]Green[/size]


[size=+2]Blue[/size]

Which of these is the best portrait? As usual, there really is no contest. Facial details live in the green channel. Red is nearly completely blown in the face and blue is too harsh and noisy. So we might be tempted just to use the green channel and be done with it, which is often the right idea with portraits. But in this case, the red channel has soething going for it: better contrast in the sky, so we'd like to get the best of both worlds: the face from the green and sky from the blue.

Theoretical digression, skip if you like. Why does the red channel have more contrast in the sky? Skys are blue and white, so you might think you'd find them in the blue channel. But the sky is almost completely uniform in the blue.

The reason is that in order for something to be either blue or white, the blue channel has to be bright. White is blue+grreen+red. If something is just blue, there will be no red. If it is white, it needs red as well as blue. So the contrast comes from the red channel.

Similarly, faces are basically red. That means that the contrast won't be found in the red channel. More green make lighter parts of faces. Less green will make the shadows.

End of digression.

It's easy to get the best of both worlds by layering the red channel over the green channel in Darken blending mode. Like this:



And the resulting image:



Here I copied the red and green channels each into their own layer and set the blending mode of the topmost layer to Darken. This blending mode uses the layer only where it is actually darker than the layer below. Since the man's face is red, it's pretty much guaranteed to be lighter in the red, and hence not used. But the sky is darkest in the red, especially the blue parts, so we get the nice contrasty sky. And it was painless, not brush work or selections. We relied only on the basic properties of the image.

Theoretical digression, skip if you like. This particular kind of blending, darken or lighten, is unique to digital manipulation. You can't do this particular thing in the darkroom, at least not easily. Filters can highlight one channel or another, but can't do this kind of logical comparison. So this is not a film look. In the darkroom, it would probably require multiple exposures with dodging and burning. Hooray for digital!

End of digression.

So far, so good. I have an action which aides to to getting to this point. Pick it up here. We're now down with channel blending. Flatten the layers and let's move on.

2. Curve to enhance contrast

Curves often seem daunting at first, but they are huge creative tool and learning to use them will give you far more control over your results than any single other PS technique you can learn.

One thing to notice here, I have my curves set to have darkness on the right, which is not the PS default. The reasons for this are not really relevant, just notice it.

So, I make a new curves adjustment layer for this image. In CS4, it shows the histogram in the background and I can instantly see that the image doesn't have a white point:



In older versions of PS, you won't have this histogram directly on the curves adjustment layer, but it's not that important. With or without the histogram, we want to find out decide what will be the lightest and darkest points of the image so we can use the full range of contrast available. This will almost always make the image look better and sharper with true rich blacks and good detail where we want it. In this case, the lightest part of the image we care about is on the man's shirt. For older versions of PS, you can discover this by holding down the Apple or Alt key and mousing over the image. You'll see the spot on the curve. For CS4, with a curve adjustment layer, you have to enter a mode to do this. There is an icon of a finger on the left side of the curves panel. Click it and then the curve will show the brightness of the point under the cursor.

Anyway, I've decided that I want s point on his shirt to be the lightest point in the image. I still want detail in the shirt, but there is nothing I care about that is brighter than the shirt. And his glasses are already black and I do care about them. So, I moved the light end of the curve inward to make the shirt point white. That's shown in the curve above, and here is the result:



This is a definite improvement, a lot less muddy, but we can do better. The punchline of this image is the man's face and the sky, not the shirt nor the glasses. So we want to use more of our contrast budget to enhance the face and sky than on the shirt and glasses. By mousing around on the curve as above, it's evident that the face and sky live in the midtones. If we get the curve to be steep there and shallower at the two ends, we'll allocate our budget appropriately.





This is an example of the classic "S-shaped" curve. Often it suffices just to pull up or down one point on the curve, depending on where we want the contrast. Another thing to notice is that fairly small changes to the curve can have dramatic impact on the image. You have to learn to be pretty subtle.

At this point, we are done with curves, time flatten the layers and to move along to the next step.

3. Sharpen two ways for local contrast and sharpness

Before we get started with this, I need to issue a warning. I only have small low resolution image to work with. That greatly influences the choice of radius to use for sharpening, so beware. I'll give more some guidelines for full sized images, but the real trick for sharpening is to look at see how it works instead of fixating on some number to use.

OK, I like to sharpen my images twice, once with a very wide radius and very low amount (called "HIRALOAM, or High Radius Low Amount" sharpening) and a second time with a much smaller radius and larger amount (called "conventional sharpening".)

HIRALOAM sharpening emphasized the large features of your image and also adds local contrast. It works even for somewhat soft or noisy images. It's something you will be happy to learn.

Make a duplicate layer and open Filter->Sharpen->Unsharp Mask. Set the Amount to its maximum value, 500 and the Threshold to 0. Now increase the Radius until you see the large features of the image emphasized but not so much that they blend into one another.





(In full sized images, the radius might be somewhere between 10 and 50.) Of course this is pretty ugly, not what we want at all, but we can see that it emphasizes the large features at the expense of detail.

So this is a lot too much of a good thing. Turn down the Amount until the obvious halos disappear, usually around 35. It might still be a little harsh looking, but we're not quite done yet.



Unsharp mask works by making light and dark halos. It surrounds dark things with light halos and dark things with light halos. You can find out a lot more about this here and here. Suffice to say, the light halos are usually a lot more noticeable than the dark ones, so it's usually a good idea to tone them down more than the dark ones. Duplicate the sharpened layer and blend one layer in Darken mode and one in Lighten mode. Then you can reduce the opacity of the Lighten layer, say to 50%, effectively dimming the light halos while keeping the dark ones.



At this point you may want to fine tune the amount of sharpening even further. It's just a matter of adjusting the opacity sliders.

In addition to emphasizing the large features of the image, it's often a good idea to do conventional sharpening to emphasize the fine details as well. In this particular image at this resolution, there's not much that's going to really change. Follow the links above for details. Here is my final version:



It's picked up some whisker detail, maybe too much. A lower amount or lighten/darken layers might look better to another eye.

pathfinder

Excellent Tutorial, John.

I am going to sit down and work this through, step by step, as soon as I get the second installment of my Journey posted.

rutt

A second example with more resolution would help. I'm on the lookout for one.

BradfordBenn

Very helpful. Going to look for stuff I can try it on...

Dogdots

Thanks for posting this -- its a big help. I'll be trying it out tomorrow

craig_d

Luminance and B&W Conversion
Nice tutorial, but there are alternatives to manual channel blending that you missed. One in particular is important because it can be used to significantly reduce digital noise.

In the two major pieces of software that I use for image processing (UFRaw for loading RAW files, and the GIMP for more complex PP), you can choose to convert to B&W either with channel blending, as you have described, or by alternate methods that preserve lightness or preserve luminance. I assume Photoshop, Lightroom, and other programs must have similar capabilities.

As many of us probably know, noise in digital images can be classified into two types: chroma noise and luminance noise. In pure chroma noise, a pixel is the right luminance but the color is off. In pure luminance noise, the color is correct but the pixel is too bright or too dark.

Now, imagine that you had an image where all of the noise was chroma noise. If you convert to B&W by preserving luminance, you are essentially just throwing out all the chroma information in the image -- including the chroma noise -- resulting in a perfectly noise-free luminance-only image.

In reality, noise in color images is pretty much always a mixture of chroma and luminance noise. In such a case, the "preserve luminance" mode of B&W conversion will quite noticeably reduce the noise level in the image, though of course the luminance noise will remain.

You might think that one disadvantage of this process is that you can't control the B&W tones the same way you can when mixing channels. When processing RAW images, though, you can still control tones through the white balance settings. When you're planning to reduce the image to B&W, it hardly matters if the WB settings are "correct" -- they can be whatever they have to be to generate the result you want.

My own B&W conversions are usually done in "preserve luminance" mode when shot at high ISO levels. I haven't tried to systematically measure the noise reduction achieved this way, but subjectively I would say that it often gives the impression of about one stop lower ISO. When I was mostly using a Canon Rebel XSi, I tried to stay at ISO 400 or less for color shots, but I could easily get good results at ISO 800 if I intended to convert to B&W.

rutt

craig_d wrote:

Nice tutorial, but there are alternatives to manual channel blending that you missed. One in particular is important because it can be used to significantly reduce digital noise.

There are many alternatives I've missed, but this is what I've been doing.

I'd very much like to learn more about your (and other approaches.) Please consider posting a detailed tutorial (probably in a new thread.) Thanks.

BTW, these days I'm learning to like a little noise now and again.

craig_d

rutt wrote:

BTW, these days I'm learning to like a little noise now and again.

Absolutely, often a touch of graininess is beneficial. But when you want it, it's easy to get, either by boosting ISO or in PP. Getting rid of it when you don't want it can be more challenging, though, and the "preserve luminance" B&W conversion is a nifty way to do it.

rutt

I have a new example, this time from Pathfinder. This is a full sized original this time, so each image I use will also be a link to a full sized version, in case you are interested.

Here is the original and a B&W version produced with my workflow:



OK, so here we go.

1. Convert to B&W by channel blending
Let's look at the RGB channels:



As expected, green is the hands down winner for best face and red is the only channel with any sky at all. So here is red layered in Darken mode over green:



So that's the end of channel blending. Flatten and move on.

2. Curve to enhance contrast

Besides setting white and black ponts, we want our curve to emphasize what sky detail we have most of all to add detail and contrast to the face of our subject. This imge is pretty much segregated into a light zone (sky) and dark zone (soldier). So we move in the endpoints to make light and dark points and get the curve steep in the highlights (sky) and midtones to shadow (soldier.)



And the result on the image:



I've been aggresive about setting a white point in the clouds. Moving the light point in a bit less might preserve a little more cloud detail in the highest highlights (but see extra credit, below.) Other than that, our goals are achieved. flatten again and on to sharpening.

3. Sharpen two ways for local contrast and sharpness

First a HIRALOAM sharpen. Here is 500% with a radius of 55.3px:



The big facial features are emphasized and the fine details are obliterated. Take the amount down to 35% and we have this:



A bit harsh, but on the right track. The light halo around him is just visible. Let's take the opacity of the light halos down by 50% with separate Lighten and Darken layers:



Better, but still too much for me. Flatten onto a duplicate layer of the background, and reduce opacity to 60%. This time I used my eye to pick the amount that looked best to me. Too much moire than 60 looked too harsh. Too much less and it looked softer.

. Flatten...

...and proceed to conventional USM. WIth Amount 500, Radius .8, and Threshold 2 I got:



I like what this has done for his eyes and buttons in particular, but not so much his uniform which is now lightened with a texture that seems wrong. What if we play the same lighten/darken trick we just used for HIRALOAM to reduce the opacity of the lightening effect of USM by 50%:



Looks pretty good to me now, so I'll call it done.


Or will I. There is one more thing we could think about here.

Extra credit -- a little more drama in the sky

Back when we had separate layers for red and blue, there is a step we might have considered. What if we curved the red layer before blending in Darken mode with the blue? We could apply quite a steep curve to the sky and so long as we kept the mid tones and shadows the same, we could still blend in darken mode and only change the sky. Perhaps a curve like so:



which makes the red layer look like this:



Notice this isn't good for the face, which is even flatter than it was before the curve. But that doesn't matter, since it's lighter than the corresponding area of the green layer, when we blend in Darken mode, we'll pick up the good green instead of the bad red:



We can now proceed as before and curve to preserve the good sky, add white/dark points, and allocate more contrast to the face:





From this point, we can sharpen 2 ways as before, but it's really no different, so I have left as an exercise for the reader.

pathfinder

The first B&W image I am posting is my original rendition, using B.D.'s Tri-X action with a bit of a curve that I created in post processing last summer after returning home from Gettysburg.

I find I can recreate it pretty closely as I said by using B.D.'s Tri X action and the following curve for this image.

B.D.'s Tri X action is --> Image>Channel Mixer> Red 43 - Blue 23 - Filter >Sharpen > Unsharp Mask -> 15, 50, 0 - then a curve as desired ( which I have recorded as an action in CS4.)

I used this curve and these sharpening settings.

[imgl]http://pathfinder.smugmug.com/photos/717487607_2aJ7F-XL.jpg[/imgl] [imgr]http://pathfinder.smugmug.com/photos/717487612_Ujk6n-XL.jpg[/imgr]

This is my new attempt at my image of the artilleryman.



As I compare it to rutts, the 'reds' are not as near to black as are his. He has chosen to render the shirt and the red band of the hat as true blacks ( my true black is in the shadow beneath his leather band of his hat) , and the greys of his vest as dark grey whereas mine are the lighter faded grey he was wearing. My reds and greys are lighter tones, as are his facial tones generally. My highlights in the sky do have ink on the paper, while rutts, will be totally white. The light was overcast and soft that day, and did not reflect the harsher tonality of rutt's image. His sky and mine approach each other in contrast.

I think the differences between our images come down to our artistic interpretatiosn of the image, not what specific tenchnique we used to get there. I like dark, dramatic images, quasi Dragan style, but did not feel that was representative of what I saw and captured that day of this artilleryman.

I told rutt I would give his technique a try so I have another image from Gettysburg to work up in progress

Dogdots

I'm finding this so interesting. Is there a way I can print this all out?

pathfinder

Mary, your browser must have a print button somewhere, doesn't it?

craig_d

pathfinder wrote:

Mary, your browser must have a print button somewhere, doesn't it?

No doubt, but it is occasionally annoying that there seems to be no way to see all of a discussion on a single page.

DavidTO

craig_d wrote:

No doubt, but it is occasionally annoying that there seems to be no way to see all of a discussion on a single page.

You'll notice in the Thread Tools pull-down menu that there is a Show Printable Version option. Once you're in the printable version there is an option to show all posts on one page.

craig_d

DavidTO wrote:

You'll notice in the Thread Tools pull-down menu that there is a Show Printable Version option. Once you're in the printable version there is an option to show all posts on one page.

Slick, I'd never noticed that. Thanks.

Dogdots

pathfinder wrote:

Mary, your browser must have a print button somewhere, doesn't it?

Yessss it does I was just looking for a way to print it without all the other stuff. I should of said that first

Dogdots

DavidTO wrote:

You'll notice in the Thread Tools pull-down menu that there is a Show Printable Version option. Once you're in the printable version there is an option to show all posts on one page.

Thanks

Taught me a lesson -- look at everything before asking

rutt

pathfinder wrote:

I think the differences between our images come down to our artistic interpretatiosn of the image, not what specific tenchnique we used to get there. I like dark, dramatic images, quasi Dragan style, but did not feel that was representative of what I saw and captured that day of this artilleryman.

It's a universal law that photographers always prefer their own versions to anyone else's. Of course,they made all the aesthetic decisions and therefore always feel that they captured the "real" scene better.

Step back a little. My technique gives wider options than workflows without the channel blending step. You get more facial detail once the red channel is eliminated and more sky detail in the red channel than the other two. You don't need to keep that detail, but you have it to work with. The cure step give you a chance to soften and to decide just how light, dark to make things. I chose to blow the whites in the sky, but of course, a different curve would have a softer sky. I chose to push face into the 3/4 tones, but that was a result of the steepness of the curve. The question really isn't whether you prefer my version to your own. It's unlikely that you will. The question is whether once you learn my technique, do you prefer the results you get with it to what you could do before.

I disagree about sky detail.

pathfinder

I agree with the advantage of knowing many ways to accomplish a task.

Your technique offers several options that can be quite useful. I will try it on a few more of my images, and post them here a bit later. BD's Tri X action is very fast and easy, and quite appealing for that reason as well. I will look for an image needing a darker, more dramatic rendering, perhaps.

rutt

pathfinder wrote:

BD's Tri X action is very fast and easy, and quite appealing for that reason as well..

Really this comparison is about how often it's worth while to spend a few extra minutes on an image instead of using canned defaults.

It takes me less than a minute to decide whether I like the blue and/or red layer blends. Then I write a curve, maybe a few minutes, then I sharpen twice, with parameters I know pretty well. Total, maybe 3-5 minutes. Of course, if you just take B.D.'s default curves and sharpening parameters, his method takes pretty much no time. But:

1. The layer blending step of my workflow is very fast. Usually, no blue channel except red clothing or other synthetic reds. Usually red layer only for sky or water in the background.
2. Lots of bang for the buck for a custom curve, but of course you could use B.D.'s canned curve to save yourself a minute or so.
3. B.D. sharpens only once, HIRALOAM. That's fine except when it isn't. You know that. B.D. uses HIRALOAM with fixed radius and amount. Again, that's fine except when it isn't. You'll sometimes get visible halos. The fixed sized radius can miss pretty badly depending on the size of the subject(s) in the field of view. Again, you know that. But, of course, you can simplify this step of my workflow by using B.D.'s canned HIRALOAM sharpening parameters.

In short, the blending step is what most distinguishes this technique from B.D.'s. My claim is that it opens up lots of potential for quality in the following steps. If I wanted a mindless B&W workflow, I'd always just take the green channel and apply some precooked curves and sharpening parameters. And I think it would have a better batting average than B.D.'s for people shots. Mindless might be good for images which you just want to show on the web or something. But is it what you want for prints of your best images? What about your first magazine cover? The smart photographer budgets post time depending on an image's purpose.

Flyinggina

Your tutorial is excellent. I like the control and the neat way to increase contrast in different areas of the photograph without using layer masks.

I'm looking forward to working through the tutorial.


Virginia