GRAPH CUT OPTIMIZATION FOR THE MUMFORD-SHAH MODEL

摘要

In this paper, we introduce a Graph Cut Based Level Set (GCBLS) formulation that incorporates graph cuts to optimize the curve evolution energy function presented earlier by Chan and Vese. We present a discrete form of the level set energy function, prove that it is graph-representable, and minimize it using graph cuts. The major advantages of this formulation include the existence of global minimum and its insensitivity to initialization. Numerical implementations show that minimizing the energy function in this non-iterative manner improves the speed of the algorithm dramatically. This makes it more appealing to real time applications such as object tracking and image guided surgery. Yet, all the advantages of using level sets methods will still be preserved.