The model used to represent documents is based on graphs for describing the form layout. We use non-hierarchical Attributed Relational Graphs (ARG). The ARGs are typically used for image representation and analysis in computer vision systems. The ARG nodes describe objects or parts of objects, while the arcs describe, using numerical attributes, the mutual relationships between the elements represented by the nodes. In particular, in the graph developed to describe the form layout (referred as a Form Graph), the nodes represent lines, instruction fields and information fields. Each arc represents the mutual position of the items corresponding to the linked nodes. Regarding the numerical arc attributes, we consider an oriented distance, by means of an attribute representing the orientation of the vector connecting the barycentres of the objects connected by the arc. By using ARGs, we have an accurate and flexible description of the form class.
In order to build the Form Graph, the user loads the image of a prototype form (Section 2.1), and insert the items present in the graph. For instance the user can insert a line( 2.2), define an instruction node ( 2.3), and the corresponding information node (2.4) with the arc connecting them (2.5).
