Integrated segmentation and recognition of handwritten numerals: comparison of classification algorithms

In integrated segmentation and recognition (ISR) of handwritten character strings, the underlying classifier is desired to be accurate in character classification and resistant to non-character patterns (also called garbage or outliers). This paper compares the performance of a number of statistical and neural classifiers in ISR. Each classifier has some variations depending on learning method: maximum likelihood estimation (MLE), discriminative learning (DL) under the minimum square error (MSE) or minimum classification error (MCE) criterion, or enhanced DL (EDL) with outlier samples. A heuristic pre-segmentation method is proposed to generate candidate cuts and character patterns. The methods were tested on the 5-digit Zip code images in CEDAR CDROM-1. The results show that training with outliers is crucial for neural classifiers in ISR. The best result was given by the learning quadratic discriminant function (LQDF) classifier.

[1]  Horst Bunke,et al.  Off-line handwritten numeral string recognition by combining segmentation-based and segmentation-free methods , 1998, Pattern Recognit..

[2]  Thomas G. Dietterich,et al.  Improving the Performance of Radial Basis Function Networks by Learning Center Locations , 1991, NIPS.

[3]  Venu Govindaraju,et al.  Probabilistic model for segmentation based word recognition with lexicon , 2001, Proceedings of Sixth International Conference on Document Analysis and Recognition.

[4]  Hiroshi Murase.  Online recognition of free-format Japanese handwritings , 1988, [1988 Proceedings] 9th International Conference on Pattern Recognition.

[5]  Biing-Hwang Juang,et al.  Discriminative learning for minimum error classification [pattern recognition] , 1992, IEEE Trans. Signal Process..

[6]  John S. Denker,et al.  Improving Rejection Performance on Handwritten Digits by Training with Rubbish , 1993, Neural Computation.

[7]  Urs Ramer,et al.  An iterative procedure for the polygonal approximation of plane curves , 1972, Comput. Graph. Image Process..

[8]  Yasuaki Nakano,et al.  Segmentation methods for character recognition: from segmentation to document structure analysis , 1992, Proc. IEEE.

[9]  Michael T. Manry,et al.  Classification-based segmentation of ZIP codes , 1993, IEEE Trans. Syst. Man Cybern..

[10]  Seong-Whan Lee,et al.  Integrated segmentation and recognition of handwritten numerals with cascade neural network , 1999, IEEE Trans. Syst. Man Cybern. Part C.

[11]  Fumitaka Kimura,et al.  Handwritten numerical recognition based on multiple algorithms , 1991, Pattern Recognit..

[12]  Masaki Nakagawa,et al.  Evaluation of prototype learning algorithms for nearest-neighbor classifier in application to handwritten character recognition , 2001, Pattern Recognit..

[13]  Fumitaka Kimura,et al.  Modified Quadratic Discriminant Functions and the Application to Chinese Character Recognition , 1987, IEEE Transactions on Pattern Analysis and Machine Intelligence.