3D Object Classification and Segmentation Methods - High-Quality Visual Experience - page 342

Fig. 3 Experimental input volumetric neurodata
with the anatomical location (voxel location) and accurate intensity values. Neu-
rodata can be segmented into three main tissue types, i.e. initial regions of interest:
grey matter, white matter and cerebro-spinal fluid. Regions of interest are set
according to intensity values.
To detect the subvolumes it is necessary to use one of the segmentation tech-
niques. This work introduces the usage of automated image segmentation tech-
niques, based on the modified Canny edge and corner detection along with global
thresholding techniques, since segmentation requires the classification of voxels
and it is done based on kernel shapes. The strategy of edge-based segmentation is
to find object boundaries and segment regions enclosed by the boundaries. Sepa-
rating head from background is accurate enough because there is a great intensity
difference between voxels associated to the skull, which represents the boundary,
and those from the background. Results of edge and corner detection are shown in
Fig. 5. As a final step of the segmentation, head voxels inside the boundary are la-
belled and prepared for classification step.