Qualitative/fuzzy approach to document recognition

Abstract

Qualitative and fuzzy representations are presented for modeling the geometric and topological attributes of orthogonal image blocks in a document layout. Qualitative models provide a compact representation since there is no absolute metric; at the same time they preserve much of the important information such as tangency relations and symmetry. A fuzzy logic approach allows some degree of uncertainty in recognizing document blocks and modeling human perception. It is shown that using qualitative reasoning in conjunction with fuzzy logic permits the production of a model that is capable of handling a wider class of documents despite variable locations and sizes of the blocks in the document. The model is based on a new point-interval representation which retains both topological and geometrical attributes of the block regions within a qualitative nonnumeric framework, and a grade of membership vector for each block.<>