Invariant Representations And Learning For Computer Vision