Multiple Predicate Learning for Document Image Understanding

Documentimageunderstandingdenotesthe recognition of semanticallyrelevant componentsin the layout extracted froma documentimage.This recognitionprocessis based on somevisual models,whosemanualspecification can be a highlydemandingtask. In order to automaticallyacquire these models, wepropose the application of machine learning techniques. In this paper, problemsraised by possible dependenciesbetweenconceptsto be learned are illustrated andsolvedwitha computationalstrategy based onthe separate-and-parallel-conquersearch. Theapproach is tested ona set of real multi-pagedocumentsprocessedby the systemWISDOM++.Newresults confirmthe validity of the proposedstrategy and showsomelimits of the machinelearningsystemusedin this work.

• Type: Conference
• Year: 2001
• Where: FLAIRS
• Authors: Floriana Esposito, Donato Malerba, Francesca A. Lisi
• Keywords: Artificial Intelligence, FLAIRS 2001, Layout Extracted Froma, Machine Learning Techniques, Semanticallyrelevant Componentsin