Why are SIFT features better descriptors than the normalized patches?
a) They are invariant to rotation and scale
b) They are sensitive to changes in lighting
c) They are computationally simpler
d) They are more robust to noise