Document image understanding of handwritten forms using rule-trees

Document image analysis and understanding (DIU) are applied in sequence to label the fields of handwritten forms so that contextual post processing can be used to improve recognition performance. We propose that efficient DIU systems for this task can be built using ambiguous regular grammars, and present results for one such system. A new method for describing these grammars is presented which seems better suited to DIU than other grammar description schemes such as Backus Naur Form or regular expressions