| 摘要 |
<p>An object of the invention is to precisely detect objects from a document image in layout analysis on the basis of a model with minimal information given to the model. Another object is to sufficiently cope with changes in position and size of the objects with a single model. A model contains data on spatial relationships of objects, data indicative for each of the objects of whether its existence on a page is mandatory or not, data for each of the objects corresponding to terminal nodes on the number of included character strings, and data for each of the objects corresponding to intermediate nodes on the number of immediate successors. From a document image are extracted not only character string regions 1 to 7 but also separators segmenting the image into objects and sub-separators treated as candidate boundaries. Labels indicative of object names are assigned to the character strings 1 to 7 by using the separators and sub-separators as constraints. <IMAGE></p> |
