I opened my previous colored shadow catcher and it was just as I described. However, I've come up with 3 new ways to do it that are incredibly precise. Meaning, you can tell the shadow catcher what color the ground is when lit and not lit, and it does exactly the right thing to adjust the image, rather than simply draw a fixed color in that spot.

Here's a demo render. I'll be posting the math and shader nodes as I have time. Unfortunately, I am currently in a very long meeting and can't post all the details at the moment.

Just as a teaser, the three techniques are:

1. Given the math behind partial transparency, calculate a negative ambient color that will properly darken the image with respect to R, G, and B. (Demo image here did it that way.)
2. Use a Refract node with IOR of 1 to "see" what's behind the shadow catcher. Blend that with a color, modulated by the shadow detector.
3. Use a Refract node with IOR of 1, but muliply that with a color, modulated by the shadow dectector.

I think #1 is fastest (no ray-tracing needed) but I suspect methods #2 or #3 will give more precise results by allowing more manipulation of the underlying image. I will experiment more and let you know.

---

Renderosity forum reply notifications are wonky. If I read a follow-up in a thread, but I don't myself reply, then notifications no longer happen AT ALL on that thread. So if I seem to be ignoring a question, that's why. (Updated September 23, 2019)