Paper:
An Evolutionary Algorithm for Optimizing Handwritten Numeral Templates Represented by Rational B-Spline Surfaces
Zheru Chi*, Zhongkang Lu*, Wan-chi Siu* and Peng-Fei Shi**
*Centre of Digital Signal Processing for Multimedia Applications Department of Electronic and Information Engineering The Hong Kong Polytechnic University Hung Hom, Kowloon, Hong Kong
**Institute of Image Processing and Pattern Recognition Shanghai Jiaotong University Shanghai, 200030, P. R. of China
To improve the reliability of a template-matching classifier for recognizing connected handwritten characters, we present an evolutionary algorithm to optimize handwritten numeral templates represented by rational Bspline surfaces of character pixel-boundary distance maps (PBDMs). Initial templates are extracted by training a feed-forward neural network. In simulation, 1,000 handwritten numeral templates (100 templates for each class) were extracted and optimized using 10,426 training samples (isolated numerals from NIST Special Database 3). A template-matching classifier using these 1,000 optimized templates rejected 90.7% of nonnumeral patterns (not included in the training set) while achieving a correct classification rate of 96.4% on independent isolated numerals.
This article is published under a Creative Commons Attribution-NoDerivatives 4.0 Internationa License.