Vol.10, No 2, 2011 pp. 125 - 140
UDC
004.032.26 004.93/’1 629.4.052.6
A COMPARISON BETWEEN TWO CHARACTER RECOGNITION APPROACHES
Lucian-Ovidiu Fedorovici
Dept. of Automation and Applied Informatics, "Politehnica" University of Timisoara, Bd. V. Parvan 2, 300223, Timisoara, Romania,
E-mail: lucian.fedorovici@aut.upt.ro
Abstract.
This paper presents the architecture of an Optical Character Recognition (OCR) technology application based on two approaches, a multilayer neural network and a Support Vector Machine (SVM) classifier using Zernike moments for feature extraction. The performance comparison of the two approaches is based on the similar layout of most of the characters that must be recognized. The comparison shows that the improvement of the processing performance can be obtained by creating classes of blobs that use geometric similarities, and doing OCR only on the representative blob from each class.
Key Words:
OCR engine, character recognition, neural networks, SVM classifier, performance improvements.
POREĐENJE IZMEĐU DVA PRISTUPA PREPOZNAVANJA KARAKTERA
Ovaj rad predstavlja arhitekturu tehnologije Optičkog Prepoznavanja Karaktera (OCR) zasnovane na dva pristupa, višeslojne nauronske mreže i Support Vector Machine (SVM) klasifikatora koji koristi Zernikove momente za izdvajanje karakteristika. Poređenje performansi dva pristupa se bazira na sličnom rasporedu većine karaktera koji treba da budu prepoznati. Poređenje pokazuje da se poboljšanje procesnih performansi može postići stvaranjem klasa bitova binarne slike koji koriste geometrijske sličnosti, i obavljanjem OCR-a samo na reprezentativnim bitovima u svakoj klasi.
Ključne reči:
OCR mehanizam, prepoznavanje karaktera, neuralne mreže, SVM klasifikator, poboljšanje performansi
