Handwriting recognition. Soft computing and probabilistic approaches (Q1408059)

This book takes a fresh look at the problem of unconstrained handwriting recognition and introduces the reader to new techniques for the recognition of written words and characters using statistical and soft computing approaches. The types of uncertainties and variations presents several algorithms that use modified hidden Markov models and Markov random field models to model the handwriting data statistically and structurally in a single framework. The book explores methods that use fuzzy logic and fuzzy sets for handwriting recognition. The effectiveness of these techniques is demostrated through extensive experimental results and real handwritten characters and words. This book presents several approaches to recognizing handwritten numerals and words based on Markov models, conditional rules, and fuzzy logic. Several new methods for extracting features are introduced in Chapter 2. Generally, feature extraction is divided into two categories: extracting features from binarized images or from gray-scale images. The first step in both categories in pre-processing to reduce noise and variations in images. Contours and skeletons are very popular in representing shapes of binarized images. In Chapter 2, a novel algorithm is presented that is able to extract features from both binary and gray-scale images using Gabor filters. A method for estimating parameters of Gabor filters is also presented, so that Gabor filters can adapt the width of line-segments to produce the desired outputs. In Chapter 5, a system using MRFs is presented to model statistical and structural information. The experimental results of this system are comparable to most recently published results. Compared with the results in Chapter 3 and 4, the recognition rate of this system is slightly lover due to the fact that the relaxation labeling algorithm may converge to a local optimum whereas the Viterbi algorithm used in Chapter 3 and Chapter 4 can quarantee the global maximum. In Chapters 7 and 8 two handwritten word recognition system based on the holastic paradigm are described.