Translation, rotation, and scale-invariant object recognition

摘要

A method for object recognition that is invariant underntranslation, rotation and scaling is addressed. The first step of thenmethod (pre-processing) takes into account the invariant properties ofnthe normalized moment of inertia and a novel coding that extractsntopological object characteristics. The second step (recognition) isnachieved by using a holographic nearest-neighbor (HHN) algorithm, innwhich vectors obtained in the pre-processing step are used as inputs tonit. The algorithm is tested in character recognition, using the 26nupper-case letters of the alphabet. Only four different orientations andnone size (for each letter) were used for training. Recognition wasntested with 17 different sizes and 14 rotations. The results arenencouraging, since we achieved 98% correct recognition. Tolerance tonboundary deformations and random noise was tested. Results for characternrecognition in “real” images of car plates are presented asnwell