The use of Radon Transform in Handwritten Arabic (Indian) Numerals Recognition

This paper describes a technique for the recognition of off-line handwritten Arabic (Indian) numerals using Radon and Fourier Transforms. Radon-Fourier-based features are used to represent Arabic digits. Nearest Mean Classifier (NMC), K-Nearest Neighbor Classifier (K-NNC), and Hidden Markov Models Classifier (HMMC) are used. Analysis using different number of projections, varying the number of Radonbased features, and the number of samples used in the training and testing of this technique is presented using the NMC and K-NNC. A database of 44 writers with 48 samples per digit each totaling 21120 samples are used for training and testing of this technique. The training and testing of the HMMC is different than that of the NMC and KNNC in its internal working and in the way data is presented to the classifier. Since the digits have equal probability the randomization of the digits is necessary in the training of the HMMC. 80% of the data was used in training and the remaining 20% in testing of the HMMC. Radon-based features are extracted from Arabic numerals and used in training and testing of the HMM. In this work we didn’t follow the general trend, in HMMC, of using sliding windows in the direction of the writing line to generate features. Instead we generated features based on the digit as a unit. Several experiments were conducted for estimating the suitable number of states for the HMM. In addition, we experimented with different number of observations per digit. The Radon-Fourier-based features proved to be simple and effective. The classification errors were analyzed. The majority of errors were due to the misclassification of digit 7 with 8 and vice versa. Hence, a second Structural Classifier is used in a cascaded (second) stage for the NMC, K-NNC, and HMMC. This stage, which is based on the structural attributes of the digits, enhanced the average overall recognition rate from 3.1% to 4.05% (Recognition rates of 98.66%, 98.33%, 97.1% for NMC, K-NNC, HMMC, respectively). Key-Words: Arabic numeral recognition, OCR, Hidden Markov Models, Handwritten Digit recognition, Nearest neighbor classifier.