A Hierarchical Image Segmentation Algorithm Based on an Observation Scale

Abstract

Hierarchical image segmentation provides a region-oriented scale-space, i.e., a set of image segmentations at different detail levels in which the segmentations at finer levels are nested with respect to those at coarser levels. Most image segmentation algorithms, such as region merging algorithms, rely on a criterion for merging that does not lead to a hierarchy. In addition, for image segmentation, the tuning of the parameters can be difficult. In this work, we propose a hierarchical graph based image segmentation relying on a criterion popularized by Felzenszwalb and Huttenlocher. Quantitative and qualitative assessments of the method on Berkeley image database shows efficiency, ease of use and robustness of our method.

Constructive Links between Some Morphological Hierarchies on Edge-Weighted Graphs

“…Evaluating the impact of mixing these techniques is left for future work. It also allows for designing new hierarchical methods derived from image predicate which are not necessarily hierarchical (see a first example in [22]). Finally, the links between the hierarchical methods presented in this paper and those based on self-dual tree of level lines [23] still need to be investigated.…”

Hierarchical Segmentations with Graphs: Quasi-flat Zones, Minimum Spanning Trees, and Saliency Maps

“…In particular, in [20], an extensive assessment based on the framework of [4] shows that the hierarchical method performs at least as well as its non-hierarchical counterpart while providing at once all the possible scales. The results of this article constitute the theoretical basis of the methods presented in the aforementioned references [12,14,35,19,21,22]. It also opens the door towards new hierarchical image analysis.…”

“…2 Top row: some images from the Berkeley database [4]. Middle row: saliency maps according to [19] developed thanks to the framework of this article. Bottom row: segmentations extracted from the hierarchies with (a) 3, (b) 18, (c) 6 and (d) 16 regions.…”

“…In [12,35,27], a resulting saliency map is computed before a possible extraction of the associated hierarchy of watersheds. In [19], a basic transformation that consists of modifying one weight on a minimum spanning tree according to some criterion is considered. The corresponding operation on the equivalent dendrogram is more difficult to design.…”

“…When this basic operation is iterated on every edge of the minimum spanning tree, one transforms a given hierarchy into another one. The technique is generic and was applied in [19,21,22] to the measures presented in [16,38,40] respectively. An in-depth exploration of one of these measures, namely the observation scale of [16] is detailed in [20].…”