Hand Drawn Document Understanding Using the Straight Line Hough Transform and Graph Matching

Josep Lladós1, Jaime López-Krahe2, Enric Martí1
1Computer Vision Center - Dep. Informàtica.
Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain
2Dep. Informatique, Lab. ai/mime.
Univ. Paris 8, Saint Denis, 93526 Paris CEDEX 02, France
josep@cvc.uab.es, lopez@ai.univ-paris8.fr, enric@cvc.uab.es


This paper presents a system to understand hand drawn architectural drawings in a CAD environment. The procedure is to identify in a floor plan the building elements, stored in a library of patterns, and their spatial relationships. The vectorized input document and the patterns to recognize are represented by attributed graphs. To recognize the patterns a such, we apply a structural approach based on subgraph isomorphism techniques. In spite of their value, graph matching techniques do not recognize adequately those building elements characterized by hatching patterns, i.e. walls. Here we focus on the recognition of hatching patterns and we develop a straight line Hough transform (SLHT) based method in order to detect regions filled in with parallel straight lines. This allows not only to recognize filling patterns, but it actually reduces the computational load associated to the subgraph isomorphism computation. The result is then that the document can be redrawn by editing all the patterns recognized.