Initial tests of the "Snowball" methodology for automatic fluid registrations, segmentations, and organ localization. In one scenario we would try to locate both kidneys on the volume, fully automatically, using only sparsely defined library of shapes.
 |
Red is a template shape approximating a kidney. Current positioning describes a step of iterations when template is moving across the volume with some substantial step (but smaller than the Snowball capture radius).
|
 |
After just few iterations of a fully automatic procedure, we end up with this localization. It has a certain match score. Hypothetically, an automatic localization algorithm might work simply traversing all possible positions with step (SX,SY,SZ)<Snowball Capture radius. Position with the highest score is chosen.
|
 |
| This is just for illustration purposes, how primitive features are constructed at each step of the iteration. |
