A novel position and shape adaptive initialization of region-based active contours in noisy images

In this paper we present a novel approach for initialization of region-based active contours. Our approach includes the adaptation of the initial contours (ICs) by varying the shape and location using higher order statistics on image data. The Hartigan's dip test and the excess mass method are used as unimodality measures for position and radii adaptation of the ICs. We compare our methods with state of the art initializations of active contours in terms of overall accuracy and initialization accuracy measure which we introduce. The proposed position and shape adaptive initialization methods outperform the state of the art initialization methods for synthetic and real images including different levels of Gaussian noise.