Active contours driven by local and global probability distributions

• We propose a novel local based signed pressure force (SPF) function.
• Both the parametric and non-parametric methods are adopted in this framework.
• A data-based prior probability is introduced to assist the detection of details.
• A global force is incorporated into this local framework to form a hybrid model.

In this paper, we propose a new local signed pressure force (SPF) function, which is defined based on the local probability distributions. According to different methods of probability density estimation, the SPF function is categorized into two classes: parametric and non-parametric SPF function. By incorporating the SPF function into a generalized geodesic active contour model, we obtain a novel local segmentation model. This model is capable of extracting the desired target, whose intensity possesses nonuniform property and boundaries suffer from fuzzyness. Moreover, a data-based prior probability is introduced to influence the signs of the SPF function, and the segmentation results appear to be more accurate with its assistance. In order to release our proposed technique from rigorous initialization, we incorporate a global force into this local framework to form a hybrid model. Experimental results on synthetic and real images demonstrate the superior performance of our methods.