How is the proposed approach different than other edge detection, segmentation methods?

Performance does not crucially depend on the tuning of parameters. Proposed approach is robust to presence of noise and texture in images.
Active contour methods require an initialization and non-linear minimization. Proposed approach is totally automatic and involves only linear minimizations.
A single IP model is able to deal with multiple boundaries. This is achieved by a generalized definition of IP curves. Intersection with z=c instead of only z=0.