IJDAR (2009) 12:97–108 DOI 10.1007/s10032-009-0084-x ORIGINAL PAPER SVM-based hierarchical architectures for handwritten Bangla character recognition Tapan Kumar Bhowmik · Pradip Ghanty · Anandarup Roy · Swapan Kumar Parui Received: 19 March 2008 / Revised: 11 November 2008 / Accepted: 26 February 2009 / Published online: 26 March 2009 © Springer-Verlag 2009 Abstract We propose support vector machine (SVM) based hierarchical classification schemes for recognition of hand- written Bangla characters. A comparative study is made among multilayer perceptron, radial basis function network and SVM classifier for this 45 class recognition problem. SVM classifier is found to outperform the other classifiers. A fusion scheme using the three classifiers is proposed which is marginally better than SVM classifier. It is observed that there are groups of characters having similar shapes. These groups are determined in two different ways on the basis of the confusion matrix obtained from SVM classifier. In the former, the groups are disjoint while they are overlapped in the latter. Another grouping scheme is proposed based on the confusion matrix obtained from neural gas algorithm. Groups are disjoint here. Three different two-stage hierarchical learn- ing architectures (HLAs) are proposed using the three group- ing schemes. An unknown character image is classified into a group in the first stage. The second stage recognizes the class within this group. Performances of the HLA schemes are found to be better than single stage classification schemes. T. K. Bhowmik Read-Ink Technologies Pvt. Ltd., Indiranagar, Bangalore 560 008, India e-mail: tkbhowmik@gmail.com P. Ghanty Praxis Softek Solutions Pvt. Ltd., Module 616, SDF Building, Sector V, Salt Lake City, Kolkata 700 091, India e-mail: pradip.ghanty@gmail.com A. Roy · S. K. Parui (B ) Computer Vision and Pattern Recognition Unit, Indian Statistical Institute, Kolkata 700 108, India e-mail: swapan@isical.ac.in A. Roy e-mail: roy.anandarup@gmail.com The HLA scheme with overlapped groups outperforms the other two HLA schemes. Keywords SVM · RBF · MLP · Handwritten character recognition · Bangla · Fusion · Grouping of classes · Hierarchical learning architectures 1 Introduction Though optical character recognition (OCR) systems for some Indian scripts are available [1], there has not been much work on recognition of handwritten Indian scripts. The pres- ent paper deals with recognition of handwritten Bangla char- acters. Bangla is the fifth most popular language in the world and the second most popular language in the Indian subcon- tinent. Bangla script has 50 basic characters (39 consonants and 11 vowels). There are more than 300 Bangla compound characters along with vowel modifiers. The present study deals with Bangla basic characters only. Most of the handwritten character recognition problems are complex and deal with a large number of classes. A lot of research effort has been made in this direction for several scripts [2, 3] and has been applied successfully to various real life applications such as postal automation, bank check verification, etc. [4, 5]. Multilayer perceptrons (MLP) and hidden Markov models (HMM) have been used for classi- fication purpose [6, 7]. Support vector machine (SVM) has not yet been used much in handwritten character recogni- tion problems. Dong et al. [8] applied SVM classifier to im- prove the performance of a handwritten Chinese character recognition system. Camastra [9] applies SVM for English handwriting recognition. Liu et al. [10] have evaluated the performance of several classifiers for handwritten numeral recognition. In [11], Günter and Bunke combine three HMM 123