I recently discovered that just about every Photoshop adjustment gives different results depending on the color space of the image.
I noticed this effect in the early 1990s, considering it the inevitable result of applying non-linear corrections to the color planes of gamma-compressed images. I focused not so much on the differences among the same corrections applied in various color spaces, but on the (color-space-dependent) color shifts. I was concerned about the problem to such an extent that I devised an algorithm to reduce the errors, and published it in this paper: “Efficient, Chromaticity-Preserving Midtone Correction for RGB Images,” at the Second Color Imaging Conference in Scottsdale, AZ, in November, 1994. Here’s the abstract
: “RGB Gamma adjustment, the standard method for control of midtone values, causes large changes in chromaticity. This paper presents an image transform algorithm that allows control of midtone values while producing chromatically-correct results. The exact algorithm requires computations only moderately more complex than those required for gamma adjustment. Approximations to the algorithm are simpler to implement than gamma adjustment, yet produce results which are more chromatically correct.”
I didn’t apply for a patent. Anybody who wanted to implement my technique was welcome to have at it. And what happened? Unless you consider Adobe’s “Luminosity” blend mode (and I have no idea of what the math for that mode is), nobody picked up the approach I suggested.
Now it’s moot, because we have so much processing power we can throw at the problem that we don’t need approximate methods. Still, one blend mode aside, there hasn’t been an attempt in Ps to reduce color shifts that arise through applying curves to RGB images. Upon consideration, I think that is A Good Thing. The reason is, over the years I have become a big fan of transparency in image processing, which I define as the user knowing exactly the processing steps that are performed on the image.
In the early days of image editing, there wasn’t a conflict between transparency and “do what I mean” transformations. Pretty much everything was transparent if the user knew a little about color science and image processing algorithms, and, if the results weren’t what the user expected, that was just tough.
Now, thanks to almost twenty five years of commercial success of desktop image editing and an incredible increase in desktop processing power, transparency is fading away. Powerful third-party plugins do their mysterious work magically -- Arthur C. Clarke said once that “Any sufficiently advanced technology is indistinguishable from magic" -- with algorithms that are secret. Lightroom started out with a “don’t worry about how we do it” mindset, and has become more opaque with each major release.
To some, that’s an unalloyed good thing. I have a more nuanced response. I like what the new plugins and the new Lightroom processing can do, but I can’t help but think that I’d be better at using the controls if I knew exactly how they worked.
That’s why I’m happy that there are still curves with normal blending in Photoshop. It’s the devil I know.