Document modeling for form class identification

Abstract

This article deals with the description of a document system analysis based on document modeling. This system is applied to forms which are used by the CAF, the French national family allowance Department -Caisse d 'Allocations Familiales. The system is composed by three different modules which deals with the different form processes. The first module - low-level processing - is divided into three stages : acquisition, binarisation and skew correction. These stages allow the transformation of a paper form into an image with correct qualities. The second module - document structuration - processes this image to extract the information contained in the form. The information is arranged to obtain a tree. This tree shows the organisation of the form content into a hierarchical way. In addition to the tree extraction, the document structuration module allows the creation of a form model base. The last module -form class identification - uses the tree and the form model base. It is composed with two pre-classifiers to extract possible lists of forms and a structural classifier. The two pre-classifiers filter the form classes among the 250 classes in order to reduce the treatment of the classifier. This classifier is based on graph matching to compare the tree of the particular form and the possible list of form extracted during the two pre-classifiers.