We have presented a new approach to object labelling in which labelling is performed in a hierarchical manner based only on relative logical layout constraints. The constraints are held in models which are encoded as rule tree data structures. Rule trees are particularly efficient at encoding these constraints and allow a variety of useful data analyses to be easily performed. Results gathered for the labelling process are encouraging and show that labelling models encoding millions of different possible labellings may be searched quickly with negligible memory requirements.
Cracknell C R W