Doug Kerr
Well-known member
This issue came up during some discussions with Kevin Stecyk.
As I understand it, when we convert from one color space to another, if the gamut of the "source" color space is not completely accommodated within the gamut of the "destination" color space, the way this is deal with varies with the "rendering intent" that is chosen. The choice of rendering intent also affects how, if at all, differences between the white points of the two color spaces are handled.
Again, as I understand it, for the three most important rendering intents, the major characteristics are:
Absolute colorimetric: No adjustment is made for differences in the white points; colors are mapped into the destination color space; colors that would map out-of-gamut in the destination color space are "clipped" to the gamut boundary.
Relative colorimetric: Adjustment is made for differences in the white points; after that, the colors are mapped to the destination color space; colors that would map out-of-gamut in the destination color space are "clipped" to the gamut boundary.
Perceptual: Adjustment is made for differences in the white points; after that, the colors are mapped into the new color space using a mapping that is "smoothly compressed" from the theoretical mapping so that the all colors from the source gamut will map "nicely" within the gamut of the destination color space.
Now, that having been said, I did some experiments in Photoshop CS5 to see this at work. I started with a file, in the Adobe RGB color space, having a red histogram with a substantial "hill" at the top end (but it went to ground before the end of the scale; there were only a few R pixels with value 255).
I then had PS convert this image into the sRGB color space (which has a smaller gamut), using different rendering intents for different trials.
I would expect that when using the perceptual intent, there would be no significant "pile-up" of R pixels at the top of the scale. But in fact there were a gigantic number of R pixels with value 255 (over 25% of all the R pixels).
What am I missing here?
Best regards,
Doug
As I understand it, when we convert from one color space to another, if the gamut of the "source" color space is not completely accommodated within the gamut of the "destination" color space, the way this is deal with varies with the "rendering intent" that is chosen. The choice of rendering intent also affects how, if at all, differences between the white points of the two color spaces are handled.
Again, as I understand it, for the three most important rendering intents, the major characteristics are:
Absolute colorimetric: No adjustment is made for differences in the white points; colors are mapped into the destination color space; colors that would map out-of-gamut in the destination color space are "clipped" to the gamut boundary.
Relative colorimetric: Adjustment is made for differences in the white points; after that, the colors are mapped to the destination color space; colors that would map out-of-gamut in the destination color space are "clipped" to the gamut boundary.
Perceptual: Adjustment is made for differences in the white points; after that, the colors are mapped into the new color space using a mapping that is "smoothly compressed" from the theoretical mapping so that the all colors from the source gamut will map "nicely" within the gamut of the destination color space.
Now, that having been said, I did some experiments in Photoshop CS5 to see this at work. I started with a file, in the Adobe RGB color space, having a red histogram with a substantial "hill" at the top end (but it went to ground before the end of the scale; there were only a few R pixels with value 255).
I then had PS convert this image into the sRGB color space (which has a smaller gamut), using different rendering intents for different trials.
I would expect that when using the perceptual intent, there would be no significant "pile-up" of R pixels at the top of the scale. But in fact there were a gigantic number of R pixels with value 255 (over 25% of all the R pixels).
What am I missing here?
Best regards,
Doug