A syntactic business form classifier

A classifier is proposed in this paper to extract structural information from business forms. The classifier is built upon existing techniques and takes advantage of the highly structured nature of forms, containing lines, boxes and text. Improvements are made to widen the scope of the classifier to handle some unexpected cases from real images. A syntactic representation is built from the detected features using their positions and lengths. The information recorded in this representation is independent of scale and displacement. A filled in form can then be compared to prerecorded blank forms to see which of these fits the best. Encouraging experimental results have been obtained