Hierarchical Image Partitioning using Combinatorial Maps

Adrian Ion (1) , Yll Haxhimusa (1) , Walter G. Kropatsch (1) , Luc Brun (2)

1 PRIP - Pattern Recognition and Image Processing Group
2 Equipe Image - Laboratoire GREYC - UMR6072

We present a hierarchical partitioning of images using a pairwise similarity function on a combinatorial map based representation. We used the idea of minimal spanning tree to fi nd region borders quickly and effortlessly in a bottom-up way, based on local differences in a color space. The result is a hierarchy of partitions with multiple resolutions suitable for further goal driven analysis. The algorithm can handle large variation and gradient intensity in images. Dual graph pyramid representations lack the explicit encoding of edge orientation around vertices i.e they lack an explicit encoding of the orientation of planes, existing in combinatorial maps. Moreover with combinatorial maps, the dual must not be explicitly represented because one map is enough to fully characterize the partition.

Domaines: Traitement des images [eess.IV]

Adrian Ion, Yll Haxhimusa, Walter G. Kropatsch, Luc Brun. Hierarchical Image Partitioning using Combinatorial Maps. CVWW 2005, 2005, Austria. 10 p. ⟨hal-00021347⟩