An Invariant Descriptor For Character Classification

MONALDI, A.C.; ROMERO, G.G.; VITULLI, D.A; BLANC, A.V.

Pucon

Congreso; IX IBEROAMERICAN OPTICS MEETING AND XII LATIN AMERICAN MEETING ON OPTICS, LASER AND APLICATIONS; 2016

CEFOP-UdeC-Concepción

Object recognition, irrespective of orientation, size orposition in an image is an ability that humans take forgranted. However, for a computer, an object that hasbeen moved, scaled or rotated in an image representsa completely different object. For a computer to rec-ognize two objects as the same or to classify differentobjects special algorithms have to be developed, whoseresponse is ro-bust to changes in scale, rotation ortranslation. Fourier-Mellin Transform emerges as an al-ternative for the design of an invariant filter [1]. Briefly,it consists in a log-polar mapping of the magnitude ofthe Fourier Transform of an image, followed by anotherFourier transform, whose amplitude is invariant underrotation and size changes. In this work, an invariantone-dimensional descriptor, called Invariant VectorialFingerprint based on the Fourier-Mellin transform isproposed. It enables alphabet character of different sizesand orientation classification. In order to recognize theobject, the Invariant Vectorial Fingerprints are evaluat-ed by means of a special designed metric, whose val-ueallows classification. Results show a good performanceof the system for letter classification.