Neural Combination of ANN and HMM for Handwritten Devanagari Numeral Recognition

In this article, a two-stage classification system for recognition of handwritten Devanagari numerals is presented. A shape feature vector computed from certain directional-view-based strokes of an input character image, has been used by both the HMM and ANN classifiers of the present recognition system. The two sets of posterior probabilities obtained from the outputs of the above two classifiers are combined by using another ANN classifier. Finally, the numeral image is classified according to the maximum score provided by the ANN of the second stage. In the proposed scheme, we achieved 92.83% recognition accuracy on the test set of a recently developed large image database[1] of handwritten isolated numerals of Devanagari, the first and third most popular language and script in India and the world respectively. This recognition result improves the previously reported[2] accuracy of 91.28% on the same data set.

[1]  M. B. Sukhaswami,et al.  Recognition of telugu characters using neural networks , 1995, Int. J. Neural Syst..

[2]  Swapan K. Parui,et al.  Shape similarity measures for open curves , 1983, Pattern Recognit. Lett..

[3]  Bidyut Baran Chaudhuri,et al.  A majority voting scheme for multiresolution recognition of handprinted numerals , 2003, Seventh International Conference on Document Analysis and Recognition, 2003. Proceedings..

[4]  Yoshio Hirose,et al.  Backpropagation algorithm which varies the number of hidden units , 1989, International 1989 Joint Conference on Neural Networks.

[5]  Ching Y. Suen,et al.  Combination of multiple classifier decisions for optical character recognition , 1997 .

[6]  Keinosuke Fukunaga,et al.  Introduction to Statistical Pattern Recognition , 1972 .

[7]  Lawrence R. Rabiner,et al.  A tutorial on hidden Markov models and selected applications in speech recognition , 1989, Proc. IEEE.

[8]  M.H. Hassoun,et al.  Fundamentals of Artificial Neural Networks , 1996, Proceedings of the IEEE.

[9]  Bidyut Baran Chaudhuri,et al.  Databases for research on recognition of handwritten characters of Indian scripts , 2005, Eighth International Conference on Document Analysis and Recognition (ICDAR'05).

[10]  Santanu Chaudhury,et al.  Devnagari numeral recognition by combining decision of multiple connectionist classifiers , 2002 .

[11]  A. Krzyżak,et al.  Methods of CombiningMultiple Classifiers and Their Application to Handwriting Recognition , 1992 .

[12]  David S. Doermann,et al.  Adaptive Hindi OCR using generalized Hausdorff image comparison , 2003, TALIP.

[13]  Seong-Whan Lee,et al.  Off-line recognition of large-set handwritten characters with multiple hidden Markov models , 1996, Pattern Recognition.

[14]  Bidyut Baran Chaudhuri,et al.  A Hybrid Scheme for Handprinted Numeral Recognition Based on a Self-Organizing Network and MLP Classifiers , 2002, Int. J. Pattern Recognit. Artif. Intell..

[15]  Ishwar K. Sethi,et al.  Machine recognition of constrained hand printed devanagari , 1977, Pattern Recognit..

[16]  Sitaram Bhagavathy,et al.  The independent components of characters are 'strokes' , 1999, Proceedings of the Fifth International Conference on Document Analysis and Recognition. ICDAR '99 (Cat. No.PR00318).

[17]  Gurpreet Singh Lehal,et al.  A Recognition System for Devnagri and English Handwritten Numerals , 2000, ICMI.