A Segmentation Scheme Based on a Multi-graph Representation: Application to Colour Cadastral Maps
Résumé
In this paper, a colour segmentation process is proposed. The novelty relies on an efficient way to introduce a priori knowledge to isolate pertinent regions. Basically, from a pixel classification stage a suitable colour model is set up. A hybrid colour space is built by choosing meaningful components from several standard colour representations. Thereafter, a segmentation algorithm is performed. The region extraction is executed by a vectorial gradient dealing with hybrid colour space. From this point, a merging mechanism is carried out. It is based on a multi-graphs data structure where each graph represents a different point of view of the region layout. Hence, merging decisions can be taken considering graph information and according to a set of applicative rules. The whole system is assessed on ancient cadastral maps and experiments tend to reveal a reliable behaviour in term of information retrieval.