How many kind of feature for region-based segmentation

I have to use region-based to classify foreground (FG) and background (BG). I read many papers about that problem. However, almost papers that I read, they often using mean feature to compare the mean square error such as

(I(x)-mean(FG)).^2>(I(x)-mean(BG)).^2=>x belong to BG 

Some authors add some condition that used statistical reigon (add sigma term). How about the other feature to discribe image region. Could you suggest to me some feature? Thank you so much