The watershed transformation for multiresolution image segmentation

Abstract

For a hyperthermia planning environment the main requirements for the following numerical computation and visualization of the temperature field are the separation of several anatomical objects. Since different contrast objects cannot be seen at the same resolution, a segmentation technique using multiresolution segmentation planes is described here. The computation of these planes is based on the technique of region growing using watersheds in digital images. First, the watershed transformation is applied on the gradient image. The resulting image is then transformed into a graph on which a region growing process is performed. The iteration of these two actions results in hierarchical segmentation planes which differ in region amount and region size. Consequently low contrast objects can be separated in the lower planes whereas high contrast objects can be extracted in the higher planes.

Autonomous learning of robust visual object detection and identification on a humanoid

“…All other pixels inside the bounding box that are marked grey indicate that their status is unknown. This pre-segmented area is sent to a non-parametric, recursive watershed segmentation algorithm [17]. Fig.…”

Evaluation of segmentation algorithms for generation of patient models in radiofrequency hyperthermia

“…In the WT, a combination of discontinuity criteria and homogeneity criteria is used (Wegner et al 1995(Wegner et al , 1996. First the grey-value range is defined within which the segmentation was to be performed.…”