Q1: If I have two buttons on a screen, one Red and one Green with no text, can the find image action differentiate between them.
Q2: If one is slightly RED, can a tolerance be specified as to "REDNESS"?
Q3: If I the two buttons on a screen, one Red and one Green have the SAME text "This button lies" in say Yellow, can the find image action differentiate between them?
Ideas, Best solutions and Best practice requested.
Try it and see! A few minutes with your favourite drawing app should be enough to find out. Play with the colours and you'll get an idea of colour differentiation and match-fuzziness relate.
Sidenote: As someone who's red/green colour blind, I (along with ~9% of humanity) will be having "strong words" with whoever designed that UI...
No.
This is really the same as Q1 -- the yellow bits (there is no "text") are the same, so can the red be differentiated from the green?
The best solution to all these problems is to badger the developer to improve the UI! As noted above a design like this will, at best, annoy a significant percentage of users and at worst could lead to disability discrimination charges! They should be making their UI as visually accessible as possible -- and the good news for you is that a visually accessible UI will be easier to image detect against.
As someone who teaches accessibility, I would also have strong words with that UI “designer”.
As I made the colours up in my head, I guess I should tell myself off for choosing the wrong colours, although I am sure I have seen a Red button with Yellow text somewhere.
My question really about how sensitive find image with text was.
Red and yellow -- generally not a problem, even for the completely colour blind. Red vs green is what I was railing against (and what, in your first post, you were trying to discriminate between).
It's image detection -- there is no text. It's all about the variation and distribution of coloured pixels, assessed as the computer "sees" them rather as we, with our innate pattern-matching capabilities, do.
IMO, and still bearing in mind that I am colour blind, good KM image detection is two parts art, one part science, one part magic 
So have a play around with different colours of button with different sizes of text in them, twiddling with fuzziness, with the "Display" option ticked so you can see the strength of match as you change the variables.
The sensitivity is controlled by the allowable fuzz.
There are various things you could do. You could search for the image in each color and base your actions on the results of the two searches.
Or if you know exactly where the button is (or can find the position by other means, such as searching for something with a fixed relative position to the buttons), you can use the Pixel condition to test for redness.