In particular,
the robustness of these methods against rotation and scaling operation is reported.
FMT and DCT methods can detect rotations of up to 10° and 5° respectively and they can not detect scaling over 10%,
while the PCA method can not detect scaling and rotation transformation at all.
Recently visual local features have become extremely popular for the tasks of object detection and recognition,
due to their robustness with respect to partial occlusion,
clutter and geometrical transformations.