HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation

Interprétation interactive de documents structurés : application à la rétroconversion de plans d'architecture manuscrits

Achraf Ghorbel 1
1 IntuiDoc - intuitive user interaction for document
Abstract : This thesis is part of the Mobisketch ANR project (http://mobisketch.irisa.fr/). This project aims to develop a generic pen-based solution for the realization of structured documents: drawings, plans... The objective is to achieve a continuum between a technical scheme in its paper form and the same document in its digital interpreted form. This continuum requires two coherent analyzers: one for the recognition phase and another for the composition /editing phase. We are interested in this thesis to the recognition analyzer. The aim of our work was to develop an interactive, incremental and generic method for the recognition of structured documents. The originality of our recognition method, named IMISketch, is to solicit the user during the analysis phase. Indeed, the analysis process is able to solicit the user if it finds ambiguity cases. Two cases of ambiguity may arise: the structural ambiguity and the shape ambiguity. The structural ambiguity is solved by the analysis system when it hesitates between two different segmentations to interpret a symbol. For example, in an architectural plan, the structural ambiguity can be raised to find the right segmentation of primitives between a wall and an opening (door, window, etc.). The shape ambiguity is raised if there are several competing hypotheses to label a symbol, such an ambiguity between a door and a window. The integration of the user in the loop recognition avoids tedious a posteriori correction of recognition errors. The recognition process is based on a separation of the analyzer and of the knowledge related to the kind of document to recognize. The a priori knowledge on the structure of the document is expressed through a visual grammatical language based on some production rules. The application of each rule is quantified by assigning a score to each hypothesis associated to a branch of the analysis tree. The grammatical description is used to drive the analysis. Our rule-based analyzer is able to compete some hypotheses of interpretation, in order to solicit the user when necessary. In addition, to limit the combinatorial, the analyzer is based on a local search context. We also implement an original process based on an hybrid exploration, guided by the grammatical description, which locally accelerates the analysis while limiting the false interpretation risk. Our interactive method has been validated on handwritten architectural plans. These plans are made of walls, three opening types and a dozen of furniture classes. This work shows that the user solicitation improves the quality of document recognition.
Complete list of metadata

Cited literature [185 references]  Display  Hide  Download

Contributor : Achraf Ghorbel Connect in order to contact the contributor
Submitted on : Friday, February 15, 2013 - 11:59:23 AM
Last modification on : Tuesday, October 19, 2021 - 11:58:55 PM
Long-term archiving on: : Sunday, April 2, 2017 - 12:53:17 AM


  • HAL Id : tel-00788832, version 1


Achraf Ghorbel. Interprétation interactive de documents structurés : application à la rétroconversion de plans d'architecture manuscrits. Informatique mobile. INSA de Rennes, 2012. Français. ⟨tel-00788832⟩



Record views


Files downloads