The texture differences are apparent, though somewhat subtle. A CPA “feature discovery” approach might work well as I mentioned before. But you could also do simple statistical difference tests (t-tests, or more likely non-parametric tests are needed) on each feature between conditions to discover the greatest differences. This testing on all features would lead to multiple comparison issues if you were using them to combine into a statistical difference test, but if you use this discovery technique on these data, and then do the actual comparison on your discovered separating features from a different set of images/data, then I think you are safe (though consult your local statistician!).
Hope that helps,