Good points so lets address them. Your calculation is a bit flawed for a simple question you have 6 possible answers - 0,1,2,3,4,5. So the bot has a 1/6 chance of correctly guessing, which is still unacceptably easy. So add a second or third question raising the possibilities up to 1/36 and 1/216 respectively. Or add more images to raise the base number up from 6 to 11 or maybe 21. Suddenly you get from 1/6 odds up to 1/9261 (20 images 3 questions). The color issue would be problematic and the only way I can see of getting around it are to avoid those questions. The answer doesn't have to be language specific since we have numeric keys however the questions would have to be in a language that the human can understand. I don't see a way around this.
The point of this thought experiment is to see if it is possible to come up with a better way of distinguishing between human and bot that isn't arduous to the human. It may not be possible but considering the combined brain power associated with this site, I doubt it.