The Zone-Based Projection Distance Feature Extraction Method for Handwritten Numeral/Mixed Numerals Recognition of Indian Scripts

Handwriting recognition has always been a challenging task in image processing and pattern recognition. India is a multi-lingual, multi-script country, where eighteen official scripts are accepted and there are over a hundred regional languages. The feature extraction method is probably the most effective method in achieving high recognition performance. In this study we proposed a zone-based feature extraction algorithm scheme for the recognition of off-line handwritten numerals of south-Indian scripts. The character centroid is computed and the character/numeral image (50×50) is further divided in to 25 equal zones (10×10). The average distance from the character centroid to the pixels present in the zone column was computed. This procedure was sequentially repeated for all the zone/grid/box columns present in the zone (10 features). This procedure was sequentially repeated for the entire zone present in the numeral image (250 features). Similarly, again the character centroid was computed and the image is further divided into 50 equal zones (5×10). The average distance from the image centroid to the pixels present in the zone was computed. This procedure was sequentially repeated for the entire zone present in the numeral image (50 features). There could be some zone/zone column that is empty of foreground pixels, then the feature value of that zone column/zone in the feature vector is zero. Finally, 300 such features were extracted for classification and recognition. The nearest neighbor, feed forward back propagation neural network and support vector machine classifiers were used for subsequent classification and recognition purposes. We obtained a recognition rate of 98.05, for Kannada numerals, 95.1 for Tamil numerals, 97.2 for Telugu numerals and 95.7 for Malayalam numerals using support vector machine.