Grouping of Handwritten Bangla Basic Characters, Numerals and Vowel Modifiers for Multilayer Classification

For better performance in multilayer or hierarchical classification of handwritten text, appropriate grouping of similar symbols is very important. Here we aim to develop a reliable grouping schema for the similar looking basic characters, numerals and vowel modifiers of Bangla language. We experimented with thickened and thinned segmented handwritten text to compare which type of image is better for which group. For classification we chose Support Vector Machine (SVM) as it outperforms other classifiers in this field. We used both "one against one" and "one against all" strategies for multiclass SVM and compared their performance.