Computational theory for interpreting handwritten text in constrained domains

摘要

This paper describes a computational theory for automatic interpretation of off-line handwritten text that is constrained in content and structure. Interpreting text is regarded as converting the textual content into a predefined symbolic representation. Content constraints are lexicons associated with syntactic categories, and known relationships (e.g., semantics, world knowledge) between phrases in different syntactic categories. Structural constraints describe the text's two-dimensional phrase layout (e.g., the phrase's position in a text line). Writing style is assumed to be unconstrained (i.e., what is normally encountered in practice).Handwritten interpretation problems in this class include determining delivery point codes from addresses, amounts from bank checks, and drug and dosage from drug prescriptions. In this paper, a computational theory, including algorithm and implementation examples, describes the problem class and a solution.