Controlled accurate searches with balloons

摘要

In this paper, we propose a system that can perform controlled, accurate search for fuzzy boundaries in noisy images. Since our system is based on the balloon model (L.D. Cohen, I. Cohen, IEEE Trans. Pattern Mach. Intell. 15 (11) (1993) 1131–1147), we propose using an adaptive inflating force instead of a constant one to make the parameters physically meaningful and easy to set. A traditional constant inflating force serves to overcome both the image forces and the intrinsic contracting force of the internal energy model. Therefore, choosing an appropriate magnitude for the inflating force is very difficult. Moreover, a chosen inflating force is not physically meaningful. The adaptive inflating force we use is controlled by two parameters, one controlling the strength of edges that a balloon is allowed to overcome and the other controlling the inflation level of the balloon. The adaptive forces are thus physically meaningful, and they can be computed systematically. Used along with a pre-processing module that we developed, our improved balloons can be controlled to extract contours with a specified strength, a property that is quite desirable.