Handwritten character recognition using fuzzy logic, neuro-fuzzy, and wavelet transform approaches

Machine recognition of general handwritten text faces a number of challenges. The primary difficulty in handwriting recognition lies in the variety of writing styles by individuals at different times and by different individuals. In order to tackle these challenges, it is essential to identify the primary and discriminating characteristic of characters, and select a set of features which are both sufficiently invariant to tolerate the great variability of handwriting, and sufficiently significant to determine which word is written. The recognition of handwriting including numerals, characters and signatures has long been an important research topic in handwritten document interpretation. Considering the engineering and psychological aspects of character recognition system, the main objective of this dissertation is to devise new approaches, incorporating human-like behavior to offer flexibility to template matching and statistical methods for off-line handwritten character recognition. Three new approaches namely, Fuzzy Logic, Neuro-Fuzzy and Wavelet Transform approaches are proposed for off-line handwritten character recognition here, and a comparison is given for their performance. Fuzzy Logic approach and Neuro-Fuzzy approach belongs to a class of structured pattern recognition, whereas Wavelet Transform approach represents statistical pattern recognition. The cognitive capability of fuzzy logic combined with the topological characteristic features of character images provide a human-like recognition system for handwritten characters. Fuzzy system's capability to directly encode structured knowledge in a flexible numerical framework is combined with the learning and memorizing characteristics of neural networks in Neuro-Fuzzy approach. By using complex transforms like discrete wavelet transform, and wavelet packet transform using best basis algorithm, reduced the number of characteristic features to be considered, which in turn reduced the complexity of the recognition system. Finally, all the three methods are compared for their performance. It was obvious that the characteristic features used were discriminating, robust and perceptually salient from biological and psychophysical view points, and described the intrinsic factors of shapes and curves that make up a cursive writing.