A novel approach of initializing region-based active contours in noisy images by means of unimodality analysis

In this paper a novel approach of initializing region-based active contours using unimodahty analysis is introduced. Two different initialization methods using dip test and excess mass test to adapt the initial radii of the active contours are developed. Both methods are compared to state of the art methods and a method which is based on kurtosis and dissimilarity (KDR). It is shown that the excess mass method outperforms KDR and state of the art methods for synthetic and real images for different noise setups with respect to the overall accuracy and Cohen's kappa of the segmentation results.