Talk Abstract:

Seminar on Industrial Problems

Closed Curves in the Analysis and Segmentation of Images

March 24, 2000

Presented by:

Karvel K. Thornber
NEC Research Institute
Princeton, New Jersey
karvel@research.nj.nec.com

570 Vincent Hall
10:10 am

It has long been known that our vision takes advantage of the closed contours which border the perimeters of every object. Yet the information processing by which such contours could be identified was unknown. Given the predominance of visual-cortex neurons which detect and process segments of edges, the question becomes, What distribution of contours does a set of edge elements characterize? To this question mathematics is ugh these edge segments are curves of least constraint, I will derive these distributions, and express them in an unexpectedly concise manner. The results greatly simplify the segmentation of contours in real images, significantly out-performing all other methods, and enable us to understand first-order transitions in illusions.

It is not hard to imagine a time when computers will utilize patterns of information processing adapted from those used by regionsof the brain. This offers us the challenge of understanding the nature of this processing independent of the means employed to effect it.

Material from IMA Talk   pdf (267KB)   postscript (586KB)

Back to 1999-2000 Seminars on Industrial Problems

Back to Industrial Programs

Back to top of page