Reducing the control by curvature can clearly be desirable for features that have high localized
curvature. This illustrates the mutual dependence between and , since low values of can
accompany low values of in regions of high localized curvature. Setting to zero would
force the snake to ignore image data and evolve under its own forces. This would be rather
farcical. The influence of is reduced in applications where the image data used is known to
be noisy. Note that one fundamental problem with a discrete version is that the final solution
can oscillate when it swaps between two sets of points which both have equally low energy.
This can be prevented by detecting the occurrence of oscillation. A further difficulty is that as
the contour becomes smaller, the number of contour points constrains the result as they cannot
be compressed into too small a space. The only solution to this is to resample the contour.
6.3.3 Complete (Kass) snake implementation
The greedy method iterates around