Use of Cognitive Reading Models in Handwriting Recognition

Written Language Recognition has great significance not only as a convenient and natural way of interacting with computers but also as an important challenge in AI. Its importance as an input interface to computers is comparable to that of speech in terms of naturalness and convenience of use. Notwithstanding the strong evidence from cognitive reading theories that suggests that text recognition is an interactive process that makes efficient use of resources at every stage, traditional handwriting recognition systems have employed a monotonically cascaded architecture. The primary goal of this proposed research is to develop a computational model of handwriting recognition that is grounded in cognitive reading theories. The model will account for several aspects of handwriting recognition that have not been addressed in the literature to date. The architecture of the word recognition model will allow it to actively seek information from earlier processing stages. Depending on the quality of the image and the difficulty of the lexicon, the feature extractor will send information on only a subset of possible features, allowing an internal decision-maker to evaluate the information and request more if needed.