Got it, Robert. I now get the embedded sRGB image.
If you would've told me it was converted to sRGB I would've already assigned that space myself. Anyway I now know something about the difference between drag/drop and Save As (on some web images) and how it strips the embedded profile. Never had this problem doing this on the majority of web images.
But now I have to ask why it is so desaturated from Andrew's original file. Is this a preview generated through Soft Proofing to a lower gamut printer using Perceptual RI?
I'm trying to figure out the purpose and usefulness of this demonstration.
Hi Tim - yes, I've learnt something too, but I'm not too sure what yet, except that cutting and pasting from a web page seems a dodgy thing to do profile-wise. I copied the image and pasted it into Photoshop and even though I have the color settings to ask when pasting and 'Preserve Embedded Profile', Photoshop assigned the working profile (Adobe RGB in this case), so the pasted image looks completely washed out.
Anyway, on to your question:
Andrew's original file is in ProPhoto as you know and the image gamut is very wide, at 255,0,0 for the saturated reds, 255,255,0 for the saturated yellows etc (in other words right to the edge of the Prophoto working space). Clearly these colors cannot be contained in Adobe RGB, let alone sRGB. But if we insist on converting to a smaller working space (or have to because we are going to display the image on the web) then with the normal Relative Colorimetric working space conversion from ProPhoto to sRGB, the colors will just get clipped to sRGB 255,0,0 for the saturated reds (and same of the other colors). That's sort of OK, but the real problem is that colors like 250,0,0 and 245,0,0 will also get clipped to 255,0,0 in sRGB. The result is that you get flat areas of color and these can look terrible.
In this image, the leftmost is in ProPhoto, the middle has been converted to sRGB using the normal Relative Colorimetric mapping and the rightmost image converted to sRGB using the Perceptual mapping. Unfortunately you will see all three converted to sRGB because I can't show it to you any other way. But you can do the conversions yourself and then at least you will be able to see the left image in Adobe RGB(ish) if you have a wide-gamut moitor, which will still clip, but not as much as sRGB. But what you can do is to use the eye-dropper to see what the color values are, so even if some colors look wrong, you can tell that IF you could display them correctly they would be OK (or not, as the case may be).
So, you don't need to take my word for it as you can check it out yourself. But if you look at the left image, the black arrow shows an area which, in ProPhoto has red at 255. Below the box the values of red drop. If you look at the middle image, which has been converted to sRGB Relative Colorimetrically, all of the colors in the box are at red=255, so that this whole area has become clipped, whereas it is not in the original image. If you look at the rightmost image which has been converted to sRGB Perceptually, only the one box has red=255, below that the red values drop. So with a Perceptual mapping we don't get the flattened areas, but we do get desaturation (if we go from a wide-gamut saturated image to a smaller-gamut workspace, as in this case).
Now if you look at the red arrows, you will see that the red ball at the left has a disc around the edge and then the color fades to white. If you could view this ball in ProPhoto there would be no disc because the colors vary uniformly from the edge to the centre. In the centre image that you can view fairly correctly as it is in sRGB, you will see a wide flat band: that's caused by the RC clipping. On the rightmost image, there is no band, the ball transitions smoothly from the edge to the centre. (So, actually, the balls in the rightmost image are visually the closest thing to balls.)
But the rightmost image will be the one that seems the most desaturated, and that is because in a perceptual mapping (as done here) all of the source space colors are squeezed in to the destination space, and this can result in a desaturated look (particularly obvious on the blue balls). But because these colors are not clipped we can at least to some extent improve them by changing their saturation and/or contrast, whereas with the middle image the damage is irreparable.
To get a better perceptual mapping that doesn't desaturate the image in the same way, you can use an 'intelligent', or on-the-fly, or image-specific mapping using ArgyllCMS (and perhaps other software) with the type of commands that I've given you in an earlier post above. The reason it's better is that the software computes the image gamut and only compresses that , not the whole of the workspace gamut. In the case of the 'Andrew's balls' image it will make little or no difference because the image is on the boundary of ProPhoto already, but it will make a difference with images that are not so crazily saturated.
What I am saying here is a matter of fact and can easily be checked out (but you have to do it yourself as it can't be shown over the web). Andrew vehemently disagrees with me and thinks I'm talking nonsense, but he's wrong.
The bottom line is that if you have to convert from one workspace to another (as we do when we go from Lightroom to web) then having the choice of a Relative or Perceptual mapping can make all the difference to how the image looks. For many images it may make no difference, but for some it will.
I hope that clarifies things a bit
Robert