Recognition of Handwritten Arabic (Indian) Numerals using Radon- Fourier-based Features

This paper describes a technique for the recognition of off-line handwritten Arabic (Indian) numerals using Radon-Fourier-based features. A two stage classification scheme is used. The Nearest Mean (NMC), K-Nearest Neighbor (K-NNC), and Hidden Markov Models (HMMC) Classifiers are used in the first stage and a Structural Classifier (SC) is used in the second stage. A database of 44 writers with 48 samples per digit each totaling 21120 samples are used for training and testing of this technique. A number of experiments are conducted to estimate the suitable number of projections and number of RadonFourier-based features using the NMC and K-NNC. In addition, several experiments are conducted for estimating the suitable number of states and observations for the HMM. These experimentally estimated parameters are used in further analysis of the different classifiers. The average overall recognition rate, after the second stage, is 98.66%, 98.33%, 97.1% using NMC, K-NNC, HMMC, respectively. The presented technique proved its effectiveness in the off-line Arabic (Indian) handwritten digit recognition. Key-Words: Arabic numeral recognition, Hidden Markov Models, Handwritten Digit recognition, Nearest Neighbor classifier