A Qualitative Profile-Based Approach to Edge Detection

Motivated by the snake approach of Terzopoulos and others, we seek to study a new class of forces that we call profile forces for detecting edges in images. Basically, an edge is regarded as a curve segment C in the image plane. At each point P on the curve segment, we take a finite cross section of the image in the transverse direction. This is the profile of the curve C at P. In the ideal case, these profiles have a ramp or ridge shape. The profile is regarded as a 1-dimensional signal F with a nominal center (or origin) denoted O. One idea is to match F to some closest ideal profile F* with center O*. This distance between O and O* constitute the profile force that tends to push O towards O*. We also investigate the idea of relative differences of profiles as another source of force.

We apply these forces to snake-like objects which automatically seek edges. An important aspect of the theory of edges is their classification. The simplest kinds of edges has a consistent profile throughout the extent of the curve. More complicated edges may have systematic variations along the curve, for instance, dotted lines. We develop such snake-like that are specialized to seek out certain classes of edges. Another aspect of our research is to develop a qualitative theory of confirmation, where we propose to be able to confirm or deny the presence of edges. It is qualitative in the sense that we try to avoid arbitrary threshholding.

Here are detail results from our investigations, in particular, in Ting-jen Yen's thesis.

For further information, please contact one of us: Chee Yap and Ting-Jen Yen.