Transworld Research Network 37/661 (2), Fort P.O. Trivandrum-695 023 Kerala, India Original Article Recent Res. Devel. Pattern Rec., 5(2013): ISBN: 978-81-7895-584-1 0. A hierarchical neuro-SVM scheme for similar character recognition in Devanagari Sushama Shelke 1 and Shaila Apte 2 1 Pune Institute of Computer Technology, Pune-411043, India 2 Rajarshi Shahu College of Engineering, Pune-411033, India Abstract. One of the major challenges in the recognition of handwritten Devanagari script is the presence of similar characters. Most of the published works assume that the training and testing database of the characters come from the same class. Without this assumption, the experimental results show that the recognition accuracy lowers mainly due to the misclassification of the characters those are similar in shape. This paper attempts to handle the recognition of handwritten Devanagari characters with efforts devoted towards improving the accuracy of similar characters. A multistage recognition scheme classifies the characters hierarchically and converts an N-class classification problem down to a 2-class classification one. The characters are first classified by neural classifier followed by a 2-class SVM classifier, forming a neuro-SVM classifier. In order to further improve the recognition performance, the characters are passed through a structural classification stage before applying to neuro-SVM classifier. Correspondence/Reprint request: Mrs. Sushama Shelke, Pune Institute of Computer Technology, Pune-411043 India. E-mail: shelke.sushama@gmail.com