Hierarchical Cell Structures for Segmentation of Voxel Images

Abstract

We compare three hierarchical structures, S15, C15, C19, that are used to steer a segmentation process in 3d voxel images. There is an important topological difference between C19 and both others that we will study. A quantitative evaluation of the quality of the three segmentation techniques based on several hundred experiments is presented. This work was founded by the BMBF under grant 01/IRC01B (research project 3D-RETISEG).

A CSC Based Classification Method for CT Bone Images

“…But unfortunately S 15 is not dense. As a consequence a CSC steered by S 15 leads to oversegmentation, see [PSW05]. A solution to that problem is the cell hierarchy C 19 that is build by cell type C 19 .…”

“…C 19 is neither plain nor saturated but dense and coverable. A comparison between the CSC steered by S 15 and C 19 was done in [PSW05]. The CSC has one parameter T that steers the similarity criterion.…”

“…A small thresh-old will widely prevent segments from containing voxels of different classes. For a low threshold the disadvantages of the S 15 -CSC compared to the C 19 -CSC, see [PSW05], are negligible. Thus we focus in this paper only on the S 15 -CSC.…”

Improved Segmentation of MR Brain Images Including Bias Field Correction Based on 3d-CSC

“…It is an effective 3D generalization of the 2D-CSC (Priese 2005). In comparison with traditional region growing techniques it does not depend on the selection of seed points.…”

3D-CSC: A General Segmentation Technique for Voxel Images with Application in Medicine

“…This will lead to an "artificial" over-segmented image. An alternative to S 15 is the C 19 structure (see figure 1c), proposed first by Sturm [4], which builds up a cell hierarchy that is dense but has more complex relationships between island levels which complicates implementation. Because of the density of the C 19 cell hierarchy a segmentation will not lead to an artificial over-segmentation.…”