Using stochastic syntactic analysis for extracting a logical structure from a document image

A method of stochastic syntactic analysis is applied to extracting the logical structure of a printed document from its physical layout and keywords indicating logical components. The document is parsed as a sentence consisting of text lines and graphic objects according to a stochastic regular grammar with attributes. By using stochastic analysis, the parser can retain possible results in order of their probability, and thus, if ambiguity occurs, it selects an optimal result more appropriately than deterministic systems. A mark up system applying the method was constructed, and 87% of the logical components of manuals and 82% of those of technical papers are correctly marked up. The rate improved to 89% when the second candidates were considered, showing the advantage of the authors´ approach over the deterministic approach