LAB for black and white

[sgwrx]

hi, i took my first shots of snow on the trees. i did a few different exposures, and wisely over exposed them 1.5 to 2 stops according to my in-camera meter. i have a 10D and used a 50mm f/1.4, shot in raw. white balance was set to cloudy day, as it is heavily overcast. i did not use a gray card to wb and i probably should have.

looking at my images in PS CS using the built in converter, they have a very purple or red tint. i somewhat expected the need to convert to black and white so i'm not too worried.

i converted one image to LAB and had only the lightness channel selected. it seemed washed out. i toggled the a channel off vs. the b channel. i really seem to like the b channel combined with the lightness channel. the red/purple tint is gone, but seems to be a little blue (maybe it's my LCD monitor).

the reason i didn't use just the lightness channel is that the L + B gave a little more contrast and this especially looked good on the gray-treetrunks.

my question is, not knowing anything about LAB, what is happening when i use L + B and is this good? for converting to black and white, where can i read more about LAB and is it even the best "auto" conversion.

thank you!
steve

[61Dynamic]

The problem is not the BW conversion method but, as you pointed to, contrast. Use curves to give contrast a boost.

IMHO, Lab is not the best method for B W conversion out there. For starters, it requires you to convert the image to Lab which is a fairly destructive process in of itself. Top that off with the fact that Lab operates in a very large color gamut, using it effectively requires 16-bit/channel processing lest you risk effects such as posterizing.

Better methods are available for BW conversion that don't require a trip into Lab.

[jdemott]

Since you shot in RAW mode, you can change the WB setting in the RAW converter until the color looks right.  (I.e., there is no need to convert to black and white simply because of the color cast, but of course you can go to black and white if that is your aesthetic preference.)  If your monitor is properly calibrated, you can make the judgment about proper white balance based on appearance alone.  Otherwise, you need to check the R, G and B values for a known neutral (gray) point in the image to be sure that R = G = B.  Adobe Camera Raw has an eyedropper tool that will let you sample color points and evaluate the effect of changing the WB setting.

There are numerous ways to convert a color image to black and white.  One of the typical ones is to convert to Lab and then use the L (lightness) channel alone.  You can of course use curves or levels adjustments on the L channel to give it more contrast and a more dramatic appearance.  When you have the L channel selected by itself you are looking at the image without any color information.  If you add the b channel you have added blue/yellow color information but you are omitting the green/magenta information of the a channel.  It is very likely that the image would therefore have a color cast (blue and/or yellow).  

If you want a black and white image (grayscale) without a color cast, then you probably want to stick with the L channel if you are using Lab because that will give you a pure black and white image.  But if you want to stay in the RGB color space, then you can use the Channel Mixer adjustment which will give you quite a bit of flexibility and control over the B/W conversion.  Just check the Monochrome box at the bottom of the dialog box and then experiment with moving the sliders.  Most people find that the best results are obtained when the sum of the values of the three sliders is approximately equal to 100.

[sgwrx]

thanks. after reviewing the posts and investigating the links, perhaps b&w is not what i like best. the blue/yellow in the lab mode made the tree trunks stand out a little more to me, so that tells me the blue channel would add that. anyway, i played around a bit and ended up using the CM to cut down the red/purple tint.

i attached to screen shots of the images at 50% size. the first of each set is color using the CM to reduce the red/purple (perhaps not enough though). the second is just plain old b&w using the CM. edit: removed top picture w/o contrast enhancement.

comments please (this is really my first official landscape/nature photo and first time trying to shoot snow so don't hold back - there are things i probably haven't even thought of!)

edit: removed image

thanks

[Tim Gray]

which is a fairly destructive process in of itself.
[a href=\"index.php?act=findpost&pid=57425\"][{POST_SNAPBACK}][/a]

First, I agree that there are better ways to get to BW.

However I think that the characterization of LAB as  "farily destructive" is a bit excessive.

I suppose it is destructive in the technical sense that 1.0000000001 is  greater than 1.0000000000.  Certainly I don't know how to make the "destruction" visible.  

Try this.  Take a RAW file, convert to 16 bits, convert to lab and back to RGB save as 16 bit tif.  Take the same RAW file convert to 16 bits, save as 16 bit tif.  The only difference between the files is that one has been converted to LAB and back.  Copy one over the other, set mode to difference and run auto levels.  There is no difference visible at any magnification.  

It's also interesting to run this expirement to actually see the degree of loss in doing jpg conversions, particularly repeated openings and resavings at quality levels of 11 or 12.  I'd be curious as to the amount the "noise" is magnified by the autolevels adjustment.

[Mark D Segal]

The foremost expert in the use of LAB is Dan Margulis. He discusses conversion to black and white in pages 110 to 114 of his new book "Photoshop LAB Color". There you will see it is not necessarily the case that using LAB is the best route to the best conversions, and moving back and forth between LAB and RGB (according to Dan) is non-destructive (this point is debated in the literature, but he has challenged his students in his seminars to convert RGB>LAB>RGB a great many times and see whether there is any visible image degradation). Most techniques involve selecting the most promising of the RGB channels and working from there. Another approach that I find works extremely well is to buy "Convert to Black and White Pro" from "themagingfactory.com" and use it on a duplicate layer of the colour image (so you can trash it easily and start again if you don't like what you got the first time).

[61Dynamic]

First, I agree that there are better ways to get to BW.

However I think that the characterization of LAB as  "farily destructive" is a bit excessive.

I suppose it is destructive in the technical sense that 1.0000000001 is  greater than 1.0000000000.  Certainly I don't know how to make the "destruction" visible. 

Try this.  Take a RAW file, convert to 16 bits, convert to lab and back to RGB save as 16 bit tif.  Take the same RAW file convert to 16 bits, save as 16 bit tif.  The only difference between the files is that one has been converted to LAB and back.  Copy one over the other, set mode to difference and run auto levels.  There is no difference visible at any magnification.   

It's also interesting to run this expirement to actually see the degree of loss in doing jpg conversions, particularly repeated openings and resavings at quality levels of 11 or 12.  I'd be curious as to the amount the "noise" is magnified by the autolevels adjustment.
[a href=\"index.php?act=findpost&pid=57435\"][{POST_SNAPBACK}][/a]

It's not at all excessive. Converting to Lab and back again is a destructive process. Image information changes. How noticeable it will be varies on many characteristics of an image such as bit-depth.

And as your example states, converting a 16-bit/channel image to Lab and back again will net little degradation. Any degradation won't be noticeable. At worst, you may see some slight color shifts in some cases. The image however is degraded. Data has been lost as is the very nature of any sort of editing one can perform. Since a 16-bit/channel image has so much data to start off with the effects aren't worth much worrying about. This is why I originally stated Lab is best used with 16-bit/channel images. Try it on a 8-bit/channel image and the bird sings a different tale.


Let's not start another Morgulis debate. Honestly, if anyone wants to read that again, they should do a forum search.    


sgwrx,
The second set looks allot better than the first with the added contrast. Your color images are definitely far too magenta. you need to add more green to the image in the raw converter (tint slider).

[sgwrx]

61dynamic, what do you think of this result? i sharpened .3 radius 200%, then applied USM 50 radius 20%. i adjusted the tint in the raw converter. here are the before and after histograms in the converter process:
edit:remove images
if i went any further towards the green, a green spike would grow up pretty much like the pink/magenta spike you see in the first example.

here is the result resized down 25%:

edit remove image

thanks

[61Dynamic]

It looks much better.

A question: is your monitor calibrated?

[sgwrx]

no.  i've played around with it and adjusted by eye only. it's some type of sony LCD which isn't very good.

It looks much better.

A question: is your monitor calibrated?
[a href=\"index.php?act=findpost&pid=57522\"][{POST_SNAPBACK}][/a]

[Mark D Segal]

sqwrx, it was astute of Daniel to ask whether you calibrate your monitor. This is one of the single most important aspects of getting accurate, predictable results. From what you describe it could be that the scope and usefulness of calibrating your monitor may be limited by the hardware itself; however, if you are planning to do a fair bit of digital photography, investment in a decent monitor capable of calibration, accompanied by an investment in a calibration package such as ColorEyes Display with the X-Rite colorimeter will pay you handsome dividends in time saved and frustration avoided.

[sgwrx]

thanks, i agree. i have casually looked at calibration tools and am very interested. i go "way back" to the late 80's and early 90's and NTSC calibration for TV's & Big Screens. i used to sell high-end CRT based stuff and was able to participate in calibration. calibration fixed many of these and leveled the playground between brands as compared to when they were taken out of the box.

i have been hesitent for two reasons, the first is i'm not sure decent results can be had with my sony LCD even if it's "any type of improvement"? second, i'm not doing anything professional at this point (getting paid). but long term i do desire calibration.

what would be interesting to me is to try and compare an image from my screen to what you all might see.  for example, the comment in response to my repost of the tint adjusted image was "much better". i don't know how such a comparison could be done. further, i'm wondering just how much i can tweak using the histogram of the raw converter or this histogram after that.

sqwrx, it was astute of Daniel to ask whether you calibrate your monitor. This is one of the single most important aspects of getting accurate, predictable results. From what you describe it could be that the scope and usefulness of calibrating your monitor may be limited by the hardware itself; however, if you are planning to do a fair bit of digital photography, investment in a decent monitor capable of calibration, accompanied by an investment in a calibration package such as ColorEyes Display with the X-Rite colorimeter will pay you handsome dividends in time saved and frustration avoided.
[a href=\"index.php?act=findpost&pid=57537\"][{POST_SNAPBACK}][/a]

[benInMA]

I agree converting to LAB for B&W conversions does not make much sense.

In the end for B&W images with no tint you should end up with either an RGB image where R=G=B for all pixels, and then convert that to monochrome to ensure that no tint can enter into the picture anywhere in the various pieces of software used to print the image.

Maybe if you have a perfectly setup and calibrated system you can print from RGB or LAB and have the image appear to have no color tint but I have not been able to do so, and simply ensuring the image is monochrome seems to work perfectly with an inkjet.   A monochrome image only uses the black ink(s) and no tint appears.

But that said using LAB should not be destructive.  8-bit and 16-bit don't really have a valid meaning when you're talking about LAB color, as unlike RGB, LAB is not represented with integers.   Once you're in LAB you should be able to do anything you want and you shouldn't see posterization as everything is happening with floating point numbers.   LAB is an absolute model so gamuts don't even mean anything in LAB either.   LAB can represent all colors so gamuts mean nothing, the only reason we have Gamuts is RGB cannot represent all colors, so the gamut defines which colors RGB is representing in a given image.  The difficulty arises because none of our output devices operate in LAB.

(Doesn't photoshop gray out the 8-bit and 16-bit menu options when you convert to LAB?)

The main issues you could have with converting back and forth from LAB would be poor algorithms (mathematical precision) used in the conversion and/or using the wrong color profiles when you convert from LAB back to RGB.

e.x. if you have the following:

1. Take an sRGB image
2. Convert to LAB
3. Convert back to Adobe RGB image

You might have a color problem in the final image

I think you are much better off just sticking with an RGB image, mixing it down to monochrome with the channel mixer, and then converting that final image to monochrome though.

[61Dynamic]

But that said using LAB should not be destructive.  8-bit and 16-bit don't really have a valid meaning when you're talking about LAB color, as unlike RGB, LAB is not represented with integers.
[a href=\"index.php?act=findpost&pid=57613\"][{POST_SNAPBACK}][/a]

This statement needs correction. Bit-depth has noting to do with color spaces. It is how much data a file contains; binary data - ones and zeros. A file with no bit-depth is a file that does not exist.

Posterizeation is still possible as posterizing occurs when there is a gap in data describing a tonal or color transition. An 8-bit/channel lab file is just as or more prone to damage when edited as a 8-bit/channel image in the ProPhoto color space.

Lab is a mathematical color model of the range of colors the human eye can perceive. It is a gamut itself by definition but it is a device independent color space.

(Doesn't photoshop gray out the 8-bit and 16-bit menu options when you convert to LAB?)

The ability to switch from 8-bit and 16-bit remains available in Lab.

[Mark D Segal]

Daniel is correct.

Putting a finer point on it, the wider the colur space one is working in, the more advantabgeous it is to use 16bit rather than 8 bit depth image files, because with 16 bit there are exponentially more levels of tonality with which to fill the space, hence the risk of banding and posterization is that much lower.

[Tim Gray]

... because with 16 bit there are exponentially more levels of tonality with which to fill the space, hence the risk of banding and posterization is that much lower.
[a href=\"index.php?act=findpost&pid=57633\"][{POST_SNAPBACK}][/a]

Assuming 16 bit - I'd say the risk of banding and posterization resulting from an RGB to LAB and back conversion is so remote as to be inconsequential.   It's certainly not a reason to avoid LAB.   (but LAB still isn't the way to do BW conversions).

[Mark D Segal]

Tim: all correct.

[benInMA]

Again guys, can any of you say for sure what Photoshops internal numerical representation is for LAB color?

My understanding is it should be floating point, not integer.

When we commonly talk about RGB & CMYK images we are talking about Integer representations.

RGB = 3 x 8 or 16 bit *integers* to describe each of the 3 channels
CMYK = 3 x 32-bit *integers* to describe each of the 4 channels

LAB = 3 x 32-bit or 64-bit *floating point* numbers to represent each of the 3 channels.

Floating point representations represent orders of magnitude more data with more precision then integer representations.

e.x. the single precision floating point # represents 10^-127 to 10^127 with 23 bits of fractional precision.

The whole reason you get color errors, posterization, etc.. when working with values as integers is any mathematical formula performed on a # which involves multiplication of an integer by a fractional # or division of any type has to round up or down to the nearest integer to store the result.

You will NOT have rounding errors and other issues which cause color issues for lab mode calculations if it is indeed using floating point.  

Color spaces define a mapping from an integer which represents a small # of values onto the infinite # of colors out there.    There is a finite # of values that can be represented and each color space/gamut defines which ones are "in" or "out" for that gamut.   With floating point representation you are working on "absolute" color, there is no need to choose which colors are "in" and "out" as you can represent them all.

The only reason gamuts & color spaces would have any meaning at all in LAB color is the monitor and printer cannot display absolute color, so anything drawn on the screen has to be converted back to integer-based RGB & given a color space so it can be drawn by the graphics device.  But for the internal calculations done to implement curves, levels, etc.. on LAB data, you will have no loss as there is no need to round any of the numbers.  They will only be rounded when they have to be converted back to RGB and this should be possible with minimal errors assuming you have not created colors outside of the color space.

Now if Photoshop is actually using integers internally to represent LAB colors, which makes no sense, and I highly doubt it is, then it would of course be full of huge errors and rounding problems, as the mathematical formulas simply for converting from RGB -> LAB involve multiplying the RGB colors by a series of fractional decimal values!

[Tim Gray]

subject to correction - my understanding is that the 32bit HDR mode is the only use of floating point in PS CS.

[benInMA]

I don't work for Adobe so I can't know for sure but just researching with google I've found several references indicating all calculations internally in photoshop are done with floating point #s in LAB mode.

All data is in LAB mode internally, and it's converted to the correct space & gamut for the screen, and the UI is set up to represent everything with integers for simplicity.

So even when you work on an 8-bit RGB image with curves, the calculations are all done internally as floating point with no loss, and you only get the posterization/rounding errors when Photoshop converts the results back to 8-bit RGB to display on the screen.  You can't do anything with the internal representation as anything you do to save it to disk or output it to any device results in it getting translated to something else.

This makes a lot of sense, it means one set of algorithms inside the code for all manipulation tasks rather then separate code for every operation in every color space.  It also explains why photoshops output is often so much better then other apps which would be working in RGB internally.  (e.x. the free program the GIMP)