Off-Line Handwritten Word Recognition Using Hidden Markov Models

Abstract

This chapter describes a system that recognizes freely handwritten words off-line. Based on Hidden Markov models (HMMs), this system is designed in the context of a real application in which the vocabulary is large but dynamically limited. Handwriting recognition is one of the most challenging tasks and exciting areas of research in computer science. The success of HMMs in speech recognition has recently led many researchers to apply them to handwriting recognition, by representing each word image as a sequence of observations. The chapter proposes an explicit segmentation-based HMM approach to recognize unconstrained handwritten words. It describes the theory of HMMs, and particularly emphasizes some variants that can enhance the standard modeling. The chapter reviews the steps of preprocessing, segmentation and feature extraction. It deals with the application of HMMs to handwritten word recognition in a dynamic vocabulary. The chapter presents the experiments performed to validate the approach. It concerns the rejection mechanism considered by the system.