ABSTRACT
The active contour model in (Kass et al. 1988) uses a local gradient based edge detector to stop the evolving curve on object boundaries. A major drawback associated with this model is that the curve has a tendency to get stuck at subtle (undesired) image features/edges. These subtle features create undesired “local minima” in energy minimization. This drawback makes the model sensitive to initialization and noise. So active contour models are typically initialized by the user with a contour near the desired object of interest to avoid local minima.
