To make the IP curve framework applicable to images without relying on prior segmentation, edge detection and edge grouping. To completely automate IP curve/surface techniques being developed at Brown University such as 3D reconstruction from occluding contours [Kongbin Kang, David B. Cooper].
To introduce fresh ideas to intermediate level salient feature extraction.

Given an image, compute IP coefficients such that the IP curve(s) represent boundaries of objects that may exist in the image.

A boundary curve detection algorithm where parts of boundaries are modelled by IP curves.
Same set of parameters work for a broad range of images.