A Soft Computing Approach to Handwritten Numeral Recognition
J.F. Baldwin, T.P. Martin, O Stylianidis · 2020
Soft computing is a key technology for the management of uncertainty, encompassing a range of techniques including fuzzy methods, neural networks and probabilistic reasoning. This chapter describes a methodology for automatic discovery and structural combination of discriminating features from data so that patterns can be effectively recognized. The methodology depends on a self-constructing self-organizing network as a preprocessor which clusters the original data in an optimal number of output classes. Handwritten character recognition systems have been proposed and implemented in a number of different ways. In order to maximize the performance of unconstrained handwritten numeral recognition, the following two aspects should be considered; one is the design of a feature extractor that produces discriminating features, and the other is the design of a classifier which has good generalization power. Neural networks have been intensively studied to achieve human-like performance in recognition and classification tasks. The self-organizing map is one of the network structures widely used in pattern recognition.