Mode Conversion Issue?

My quickest and most concise question to date:

If you open an image in RGB, then convert to LAB to extract a channel, then convert back to RGB; has any "damage" been done? Assume that there are no other layers or adjustments stacked on top of the base image--it's just one background/original layer. Or should the original document be duplicated in order to do this?

No. To verify place 4 samplers in your image and note the values. Convert to LAB, duplicate your channel, then convert back to RGB. Values of sample points should be the same.

D Thompson wrote in post #16001486
No. To verify place 4 samplers in your image and note the values. Convert to LAB, duplicate your channel, then convert back to RGB. Values of sample points should be the same.

Thanks!

D Thompson wrote in post #16001486
No. To verify place 4 samplers in your image and note the values. Convert to LAB, duplicate your channel, then convert back to RGB. Values of sample points should be the same.

But what happens to the histogram? Switching between color modes introduces quantization errors and data loss.

PixelMagic wrote in post #16001559
But what happens to the histogram? Switching between color modes introduces quantization errors and data loss.

I guess the key word in the OP is "any" damage. I was strictly going by the color sample points showing no shift. Looking at the histogram you might see a very slight change in a couple of places, very minimal at least in the couple of images I checked. I doubt it would be noticeable by most, but you are correct.

@Randman - after PixelMagic's post and for no "damage" to the original you should duplicate the image, convert to LAB, duplicate your channel to the original, and delete the copy.

D Thompson wrote in post #16001691
I guess the key word in the OP is "any" damage. I was strictly going by the color sample points showing no shift. Looking at the histogram you might see a very slight change in a couple of places, very minimal at least in the couple of images I checked. I doubt it would be noticeable by most, but you are correct.

@Randman - after PixelMagic's post and for no "damage" to the original you should duplicate the image, convert to LAB, duplicate your channel to the original, and delete the copy.

Check. Thank you

The change is extremely slight and can be ignored but there is some data loss. I recall an epic argument over this issue among several Photoshop luminaries like Andrew Rodney and Dan Margulis a few years ago; I wonder if I can find it.

D Thompson wrote in post #16001691
I guess the key word in the OP is "any" damage. I was strictly going by the color sample points showing no shift. Looking at the histogram you might see a very slight change in a couple of places, very minimal at least in the couple of images I checked. I doubt it would be noticeable by most, but you are correct.

@Randman - after PixelMagic's post and for no "damage" to the original you should duplicate the image, convert to LAB, duplicate your channel to the original, and delete the copy.

The "epic" arguments involving LAB colorspace (and 16bit vs. 8bit) occurred between Dan Margulis and Bruce Lindbloom, back in 2007-2008, as I recall. Margulis is the proponent, Lindbloom is the opponent. Margulis used test images that showed insignificant damage in LAB conversions. Lindbloom used test images that did show significant damage. Much of what they wrote was beyond the comprehension of the average user.

Key points (as best I can remember):

The degree of damage from an rgb to LAB to rgb conversion depends on several variables:

1. The bit depth of the image
2. The rgb colospace used (sRGB, Adobe98, Prophoto).
3. The actual content of the image - how much the colors in the image stretch the gamuts used.
4. The number of times the round-trip conversion is made.

A better way to demonstrate the issue for yourself. Better than just a visual evaluation or using sample points:

1. Load an "rgb" image and make a duplicate.
2. Convert the duplicate to LAB and back to rgb.
3. Copy the duplicate and paste it as a new layer on top of the original.
4. Put that layer in "Difference" mode and display the histogram
(4a. Click on the exclamation point icon in the histogram to make sure you get a true histogram.)
5. Look at the "StdDev" value.

The StdDev value is the best measurement of the difference between two images. You don't have to understand the math, just know that the higher the value the greater the difference. Values less than 0.50 are generally insignificant and invisible. As the StdDev value rises above 0.50 the difference becomes more significant and visible. To better "see" the difference, add a Threshold layer on top and move the threshold slider to the far left (value 1). Then slowly increase it to 2, 3, 4, etc.

Then repeat the test, but make 20 to 30 round-trip conversions on the duplicate image before copying it back to the original. See the difference increase.

Do the test with a 16bit image in Prophoto vs. an 8bit image in sRGB. Do the test with an image with low saturation (well inside all gamuts) vs. an image with high saturation.

Bottom Line:

You can get away with one round-trip conversion 99% of the time. But you don't really need to bother because there are ways to reach the same goals with other techniques that stay in rgb mode.

Here is the only bookmark I have left that gives a good overview: http://www.luminous-landscape.com/forum/in​dex.php?topic=28753

The argument I was referring to is this one; it preceded the one involving Bruce Lindbloom by several years: Dan Margulis Applied Color Theory - Does Conversion to LAB Damage Files?

It is also easy enough to make a duplicate image and convert it to LAB and leave it open, in case you need access to the other channels, etc. There is also an extension for PS called "Channels Power Tool" that will permit you to preview and select and apply (to a layer or mask or channel of your working image) any R, G, B, L, A, B, C, M, Y or K channel for your working image. Without converting it to the desired color mode.

http://bigano.com …ol/273-channels-tool.html

it is really handy, especially if you use channels for masks or blend channels, using apply image.

You can also use this method, a variation of the Difference exercise above, to compare an original RGB image with one converted to LAB and back:

http://digitaldog.net/​files/Apply_Image.pdf

Courtesy of Andrew Rodney, aka DIgital Dog.

kirk