..............................
..............................
..............................
Recognition of Handwritten Characters Based on
This paper is devoted to the off-line handwritten character recognition based on the two dimensional wavelet
transform and a single support vector machine classifier. The wavelet transform provides a representation of the image in
independent frequency bands. It performs a local analysis to characterize images of characters in time and scale space. The
wavelet transform provides at each level of decomposition four sub-images: a smooth or approximation sub-image and three
detail sub-images. In handwritten character recognition, the wavelet transform has received more attention and its
performance is related not only to the use of the type of wavelet but also to the type of a sub-image used to provide features.
Our objective here is thus to study these two previous points by conducting several tests using several wavelet families and
several combinational features derived from sub-images. They show that the symlet wavelet of order 8 is the most efficient and
the features derived from the approximation sub-image allow the best discrimination between the handwritten digits.
Training set Testing set R.Rate SVM rbf Our paper 60000 10000 98.76% SVM rbf Our paper 50000 10000 98.60% MLP
[2] 50000 10000 97.57% SVM
[17] 60000 10000 98.09% MLP
[18] 60000 10000 98.22% SVM
[20] 60000 10000 89.64% SVMs
[20] 60000 10000 99.32% 4. Conclusion In this paper, we have presented a technique of handwritten character recognition which combines a wavelet transform and a single support vector machine classifier. The wavelet transform allows us characterizing the character images by a set of features. The relevance of these features depends strongly on the choice of the type of the wavelet and sub-images derived from the wavelet transform. In this paper, several tests including several wavelets and smooth and details sub-images derived from the wavelets are conducted in order to determine the best wavelet and the best sub-image in the handwritten recognition framework. Experimental results on MNIST database reveal that sym8 wavelet outperforms other types of wavelets as those used in the previous works. They show also that features extracted from the smooth sub- image allowed achieving the best recognition rate. The proposed technique is efficient in comparison with other handwritten recognition methods published in the literature. As future work, we intend on one hand to integrate a normalization operation as preprocessing procedure in order to regulate the position and shape of character images, so as to reduce shape variation between the images of same class. On other hand, to investigate other features extracted from the four sub- images. References
[1] Berthiaume V. and Cheriet M., Handwritten Digit Recognition by Fourier-Packet Descriptors, Electronic Letters on Computer Vision and Image Analysis, vol. 11, no. 1, pp. 68- 76, 2012.
[2] Bhattacharya U. and Chaudhuri B., A Majority Voting Scheme for Multiresolution Recognition of Handprinted Numerals, in Proceedings of the 17th International Conference on Document Analysis and Recognition, Edinburgh, pp. 16-20, 2003.
[3] Burges C. and Scholkopf B., Improving the Accuracy and Speed of Support Vector Machines, in Proceedings of the 9th International Conference on Neural Information Processing Systems, Denver, pp. 375-381, 1997.
[4] Cortes C. and Vapnik V., Support-Vector Networks, Machine Learning, vol. 20, no. 3, pp. 273-297, 1995.
[5] Gaceb D., Eglin V., and Lebourgeois F., A New Mixed Binarization Method Used in a Real Time Application of Automatic Business Document and Postal Mail Sorting, The International Arab Journal of Information Technology, vol. 10, no. 2, pp. 179-188, 2013.
[6] Gorgevik D. and Cakmakov D., An Efficient Three-Stage Classifier for Handwritten Digit Recognition, in Proceedings of the 17th International Conference on Pattern Recognition, Cambridge, pp. 507-510, 2004.
[7] Kim K., Park J., Song Y., Kim I., and Suen C., Recognition of Handwritten Numerals Using a Combined Classifier with Hybrid Features, in Proceedings of International Workshops on Statistical Techniques in Pattern Recognition and Structural and Syntactic Pattern Recognition, Lisbon, pp. 992-1000, 2004.
[8] Lauer F., Suen C., and Bloch G., A Trainable Feature Extractor for Handwritten Digit Recognition, Pattern Recognition, vol. 40, no. 6, pp. 1816-1824, 2007.
[9] Lecun Y., The MNIST Database of Handwritten Digits, http://yann.lecun.com/exdb/mnist, Last Visited, 2010.
[10] Liu C., Nakashima K., Sako H., and Fujisawa H., Handwritten Digit Recognition: Investigation of Normalization and Feature Extraction Techniques, Pattern Recognition, vol. 37, no. 2, pp. 265-279, 2004. Recognition of Handwritten Characters Based on Wavelet Transform and SVM Classifier 1087
[11] Ma J., Zhao Y., and Ahalt S., OSU SVM Classifier Matlab Toolbox, Ohio State University, USA.
[12] Majumdar A., Bangla Basic Character Recognition Using Digital Curvelet Transform, Journal of Pattern Recognition Research1, vol. 2, no. 1, pp. 17-26, 2007.
[13] Mallat S., Multifrequency Channel Decompositions of Images and Wavelet Models, IEEE Transactions on Acoustics, Speech and Signals Processing, vol. 37, no. 12, pp. 2091- 2110, 1989.
[14] Niu X. and Suen C., A Novel Hybrid CNN- SVM Classifier for Recognizing Handwritten Digits, Pattern Recognition, vol. 45, no. 4, pp. 1318-1325, 2012.
[15] Pfister M., Behnke S., and Rojas R., Recognition of Handwritten ZIP Codes in a Real-World Non-Standard-Letter Sorting System, Applied Intelligence, vol. 12, no. 1-2, pp. 95-114, 2000.
[16] Rajput G. and Mali S., Fourier Descriptor Based Isolated Marathi Handwritten Numeral Recognition, International Journal of Computer Applications, vol. 3, no. 4, pp. 9-13, 2010.
[17] Rehman A., Gao Y., Wang J., and Wang Z., Image Classification Based on Complex Wavelet Structural Similarity, in Proceedings of International Conference on Image Processing, Brussels, pp. 1-9, 2011.
[18] Romero D., Ruedin A., and Seijas L., Wavelet- Based Feature Extraction for Handwritten Numerals, in Proceedings of International Conference on Image Analysis and Processing, Vietri sul Mare, pp. 374-383, 2009.
[19] Ryszard S., Feature Extraction of Gray-Scale Handwritten Characters Using Gabor Filters and Zernike Moments, in Proceedings of Computer Recognition Systems 2, Berlin, pp. 340-347, 2007.
[20] Seijas L. and Segura E., A Wavelet-Based Descriptor for Handwritten Numeral Classification, in Proceedings of International Conference on Frontiers in Handwriting Recognition, Bari, pp. 653-658, 2012.
[21] Shelke S. and Apte S., A Multistage Handwritten Marathi Compound Character Recognition Scheme Using Neural Networks and Wavelet Features, International Journal of Signal Processing, Image Processing and Pattern Recognition, vol. 4, no. 1, pp. 81-93, 2011.
[22] Trier O., Jain A., and Taxt T., Feature Extraction Methods for Character Recognition-A Survey, Pattern Recognition, vol. 29, no. 4, pp. 641-662, 1996.
[23] Yang L., Suen C., Bui T., and Zhang P., Discrimination of Similar Handwritten Numerals Based on Invariant Curvature Features, Pattern Recognition Society, vol. 38, no. 7, pp. 947- 963, 2005. Malika Ait Aider She is a permanent research scientist in the Automatic Control Department of Electrical Engeneering and Computer Science Faculty at the University Mouloud Mammeri of Tizi Ouzou, Algeria. Her research activity is mainly focused on feature extraction and classification for character recognition. Kamal Hammouche He is currently Professor in the Automatic Control Department of the Electrical Engineering and Computer Science Faculty at the University Mouloud Mammeri of Tizi Ouzou, Algeria. His research interests are in the area of image processing and pattern recognition. Djamel Gaceb He has a PhD in computer science from INSA of Lyon. Currently he is working on the topic of business document image recognition, retrieval and analysis, industrial vision image processing on mobile devices and real time applications.
[2] 50000 10000 97.57% SVM
[17] 60000 10000 98.09% MLP
[18] 60000 10000 98.22% SVM
[20] 60000 10000 89.64% SVMs
[20] 60000 10000 99.32% 4. Conclusion In this paper, we have presented a technique of handwritten character recognition which combines a wavelet transform and a single support vector machine classifier. The wavelet transform allows us characterizing the character images by a set of features. The relevance of these features depends strongly on the choice of the type of the wavelet and sub-images derived from the wavelet transform. In this paper, several tests including several wavelets and smooth and details sub-images derived from the wavelets are conducted in order to determine the best wavelet and the best sub-image in the handwritten recognition framework. Experimental results on MNIST database reveal that sym8 wavelet outperforms other types of wavelets as those used in the previous works. They show also that features extracted from the smooth sub- image allowed achieving the best recognition rate. The proposed technique is efficient in comparison with other handwritten recognition methods published in the literature. As future work, we intend on one hand to integrate a normalization operation as preprocessing procedure in order to regulate the position and shape of character images, so as to reduce shape variation between the images of same class. On other hand, to investigate other features extracted from the four sub- images. References
[1] Berthiaume V. and Cheriet M., Handwritten Digit Recognition by Fourier-Packet Descriptors, Electronic Letters on Computer Vision and Image Analysis, vol. 11, no. 1, pp. 68- 76, 2012.
[2] Bhattacharya U. and Chaudhuri B., A Majority Voting Scheme for Multiresolution Recognition of Handprinted Numerals, in Proceedings of the 17th International Conference on Document Analysis and Recognition, Edinburgh, pp. 16-20, 2003.
[3] Burges C. and Scholkopf B., Improving the Accuracy and Speed of Support Vector Machines, in Proceedings of the 9th International Conference on Neural Information Processing Systems, Denver, pp. 375-381, 1997.
[4] Cortes C. and Vapnik V., Support-Vector Networks, Machine Learning, vol. 20, no. 3, pp. 273-297, 1995.
[5] Gaceb D., Eglin V., and Lebourgeois F., A New Mixed Binarization Method Used in a Real Time Application of Automatic Business Document and Postal Mail Sorting, The International Arab Journal of Information Technology, vol. 10, no. 2, pp. 179-188, 2013.
[6] Gorgevik D. and Cakmakov D., An Efficient Three-Stage Classifier for Handwritten Digit Recognition, in Proceedings of the 17th International Conference on Pattern Recognition, Cambridge, pp. 507-510, 2004.
[7] Kim K., Park J., Song Y., Kim I., and Suen C., Recognition of Handwritten Numerals Using a Combined Classifier with Hybrid Features, in Proceedings of International Workshops on Statistical Techniques in Pattern Recognition and Structural and Syntactic Pattern Recognition, Lisbon, pp. 992-1000, 2004.
[8] Lauer F., Suen C., and Bloch G., A Trainable Feature Extractor for Handwritten Digit Recognition, Pattern Recognition, vol. 40, no. 6, pp. 1816-1824, 2007.
[9] Lecun Y., The MNIST Database of Handwritten Digits, http://yann.lecun.com/exdb/mnist, Last Visited, 2010.
[10] Liu C., Nakashima K., Sako H., and Fujisawa H., Handwritten Digit Recognition: Investigation of Normalization and Feature Extraction Techniques, Pattern Recognition, vol. 37, no. 2, pp. 265-279, 2004. Recognition of Handwritten Characters Based on Wavelet Transform and SVM Classifier 1087
[11] Ma J., Zhao Y., and Ahalt S., OSU SVM Classifier Matlab Toolbox, Ohio State University, USA.
[12] Majumdar A., Bangla Basic Character Recognition Using Digital Curvelet Transform, Journal of Pattern Recognition Research1, vol. 2, no. 1, pp. 17-26, 2007.
[13] Mallat S., Multifrequency Channel Decompositions of Images and Wavelet Models, IEEE Transactions on Acoustics, Speech and Signals Processing, vol. 37, no. 12, pp. 2091- 2110, 1989.
[14] Niu X. and Suen C., A Novel Hybrid CNN- SVM Classifier for Recognizing Handwritten Digits, Pattern Recognition, vol. 45, no. 4, pp. 1318-1325, 2012.
[15] Pfister M., Behnke S., and Rojas R., Recognition of Handwritten ZIP Codes in a Real-World Non-Standard-Letter Sorting System, Applied Intelligence, vol. 12, no. 1-2, pp. 95-114, 2000.
[16] Rajput G. and Mali S., Fourier Descriptor Based Isolated Marathi Handwritten Numeral Recognition, International Journal of Computer Applications, vol. 3, no. 4, pp. 9-13, 2010.
[17] Rehman A., Gao Y., Wang J., and Wang Z., Image Classification Based on Complex Wavelet Structural Similarity, in Proceedings of International Conference on Image Processing, Brussels, pp. 1-9, 2011.
[18] Romero D., Ruedin A., and Seijas L., Wavelet- Based Feature Extraction for Handwritten Numerals, in Proceedings of International Conference on Image Analysis and Processing, Vietri sul Mare, pp. 374-383, 2009.
[19] Ryszard S., Feature Extraction of Gray-Scale Handwritten Characters Using Gabor Filters and Zernike Moments, in Proceedings of Computer Recognition Systems 2, Berlin, pp. 340-347, 2007.
[20] Seijas L. and Segura E., A Wavelet-Based Descriptor for Handwritten Numeral Classification, in Proceedings of International Conference on Frontiers in Handwriting Recognition, Bari, pp. 653-658, 2012.
[21] Shelke S. and Apte S., A Multistage Handwritten Marathi Compound Character Recognition Scheme Using Neural Networks and Wavelet Features, International Journal of Signal Processing, Image Processing and Pattern Recognition, vol. 4, no. 1, pp. 81-93, 2011.
[22] Trier O., Jain A., and Taxt T., Feature Extraction Methods for Character Recognition-A Survey, Pattern Recognition, vol. 29, no. 4, pp. 641-662, 1996.
[23] Yang L., Suen C., Bui T., and Zhang P., Discrimination of Similar Handwritten Numerals Based on Invariant Curvature Features, Pattern Recognition Society, vol. 38, no. 7, pp. 947- 963, 2005. Malika Ait Aider She is a permanent research scientist in the Automatic Control Department of Electrical Engeneering and Computer Science Faculty at the University Mouloud Mammeri of Tizi Ouzou, Algeria. Her research activity is mainly focused on feature extraction and classification for character recognition. Kamal Hammouche He is currently Professor in the Automatic Control Department of the Electrical Engineering and Computer Science Faculty at the University Mouloud Mammeri of Tizi Ouzou, Algeria. His research interests are in the area of image processing and pattern recognition. Djamel Gaceb He has a PhD in computer science from INSA of Lyon. Currently he is working on the topic of business document image recognition, retrieval and analysis, industrial vision image processing on mobile devices and real time applications.