Isolated Handwritten Roman Numerals Recognition using Dynamic Programming, Na�e Bayes and Support Vectors Machines

Optical character recognition is undoubtedly considered as a one of the most active and dynamic fields of pattern recognition and artificial intelligence; it really provides in fact a solution for recognizing large volume of patterns automatically. The purpose of the present study is to compare in one hand between the performances of three novel hybrid methods used in OCR for extracting efficiently the features from characters which are the structural method called zoning combined in first time with Krawtchouk, then in second time with pseudo-Zernike invariant moments then finally combined with invariant analytical Fourier-Mellin transform in third time, and between the precision of three classifiers which the first one is a statistical that is the support vectors machine, the second is a probabilistic that is the naïve Bayes while the third forms part from optimization that is the dynamic programming on the other hand. For this purpose, we have preprocessed each numeral image by the median filter, the thresholding, the centering and the edge detection techniques. Moreover, the experiments that we have applied provided us convincing and satisfactory results. General Terms Zoning method, Krawtchouk invariant moment, Pseudo Zernike moment, Invariant analytical Fourier-Mellin transform, Dynamic programming, Naïve Bayes classifier, Support vectors machine.