Knowledge-Based Graphics Understanding And Description For Document Archival And Retrieval
Sing‐Tze Bow, Wassim El Masri · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1988
Automating the input of mixed text/graphic documents require more than just a character recognition system. We require algorithms to separate text strings from graphic and also to recognize the graphic and generate a description file for it. In this paper recent results on the recognition and structural description of graphics are reported. During machine recognition on the graphics, some heuristics are introduced to equip the system with a certain amount of decision making functions so as to narrow and optimize the search. Algorithms are designed for automating the generation of loops with minimum redundancy from bit-map, identifying those loops thus generated if they are simple ones, decomposing the complex loops into simpler interpretable shapes, and finally establishing succinct description files for the graphics. Error corrections on misalignments introduced by the feeding mechanism have been given consideration. Extensive experiments have been done on various graphics, and satisfactory results have been obtained. This technique is also useful for the analysis of computer vision segmented images.