A multilayer recognition method for understanding table-form documents
Toyohide Watanabe, Qin Luo · International Journal of Imaging Systems and Technology · 1996
The items in documents are composed under the mutual relationship between geometrical layout structure and logical structure. Thus, currently used documents may be categorized roughly on the basis of the geometrical layout structure and logical structure. Table-form documents are strictly defined by the geometrical layout structure in comparison with the other documents. Since the item fields are, in general, surrounded with vertical and horizontal line segments, with a view to understanding table-form documents it is better to recognize these line segments first and then identify individual item fields on the basis of the recognition results. At least, in table-form documents the vertical and horizontal line segments take important roles in recognizing the layout structures. In this article, we address a multilayer recognition method for understanding table-form documents. Our recognition layers are composed hierarchically of four different recognition processes: document class recognition, layout recognition, item recognition, and character recognition. In addition, our recognition method is organized on the basis of the interaction paradigm that the lower-layer recognition process verifies objects interpreted in the upper-layer recognition process with its own knowledge and then decomposes the objects into more elementary and meaningful objects. Moreover, we discuss a knowledge representation method from the viewpoints of physical and logical representations, and syntactic and semantic information. We also show the knowledge that is useful in understanding table-form documents with respect to such a representation method. © 1996 John Wiley & Sons, Inc.