Understanding mathematical expressions from document images

This paper proposes a system that understands mathemaiical ezpressions on binarized printed docu-med images. The system first e&acts a set of ‘prim-dives from the image. Each of the e&acted primi-tives ti associated with a ‘bounding boz ’ and its label. Using the ai%ributes of ihe primitives, the system con-structs an initial hierarchy. Construction of an initial hierarchy includes merging a group of primitives into a key word. iVezt, the system checks the validity of the hierarchy according to conventional mathematical syntaz rules. If any syntaz error is detected, the sys-tem makes attempts to correct the errors. The modi& cation step includes reconfiguring the initial hierarchy, reviding the original image for possible missing prim-itives, placing dummy primitives into missing spots in the hierarchy, and so on. The corrected hierarchical structure can be converted into the format for a par-ticular publication system such as T@.