Quasi-automatic initialization for parametric active contours

Active contour is a well-known image segmentation technique, commonly used to find object boundaries in images. Its main benefit is its ability to retrieve an ordered collection of points. However, fitting precisely a deformable contour to actual boundaries depends strongly on its initialization and requires adjusting various parameters. This paper presents an original method to initialize quasi-automatically explicit deformable models when segmenting regions that require no change of topology. The proposed method relies on a careful study of the gradient vector flow. Two original concepts are introduced, namely strong and weak divergence centers. The analysis of the properties of these centers leads to establishing a quasi-automatic method to setup an initial curve that will reach all the boundaries of a target region. Results using synthetic and real images are presented, showing the validity of our approach.