Alan Browne
On 2014.04.19, 09:59 , nospam wrote:

nospam
the conversions are also not lossless, something which is trivial to prove. make the conversion and subtract from the original. if they're identical, the result will be zero, which it definitely is not, and on an image i randomly picked, it's noticeable without subtracting.

Alan Browne
I just did this on a high key light image. See these 4 images.

[1] Original (now in .jpf (JPEG2000) to save space) (aka: the Lab copy) https://www.dropbox.com/s/esuc08yizhndmvd/HugoBossBeltBuckle_20140323_0002.jpf

[2 Original ( .jpg to save space) (aka: the RGB copy) https://www.dropbox.com/s/i2ni8bpm738y9ej/HugoBossBeltBuckle_20140323_0002%20copy%20copy.jpg

[3] The difference (substraction - in jpg) (aka: nospam is wrong) https://www.dropbox.com/s/yuum3sfit6e5bp1/HugoBossBeltBuckle_20140323_0002-D.jpg

[4] The difference (with sharpening on the Lab copy, jpg) (aka: test that difference works). https://www.dropbox.com/s/3uwyuwun56nc370/HugoBossBeltBuckle_20140323_0002-SD.jpg

Procedure:

-Image was loaded as raw and duplicated to a 2nd image. -First image was changed to Lab -First image was saved as TIFF (from Lab 'space') ([1] above) -2nd image was saved as TIFF (from RGB 'space') ([2] above) -Both images were re-loaded (they loaded as Lab and RGB - just as they were saved). -Copied the 2nd image and added it as a layer over the first. -Difference would not work when one layer was in Lab and the other in RGB. -Converted the first image back to RGB, then replaced the 2nd image as a layer again.

*Difference was pure black (no differences - [3] above) ========================================================

-Sharpened the 2nd image to verify that differences would pop out (they did) and replaced the layer over the 1st image with it. *The sharpening difference showed ([4] above) ========================================================

So not only were the differences invisible to the eye they were NOT AT ALL shown by differencing.

Of course you're welcome to show differently.