Contour extraction by multi-level active contour model

In this paper, an innovative algorithm for object segmentation and contour extraction is proposed, where the active contour evolution based on the Mumford-Shah model is performed on a coarse-to-fine approach spanned by wavelet transform. The multi-level active contour model consists of three main parts: 1). wavelet decomposition for obtaining multi-scale image; 2) in the top-level, image wavelet-based edge detection to get an initial evolving contour (initialization procedure); 3) evolving contour based on the Mumford-Shah model in each level, from top level to down level. The experiments and analysis demonstrate that the whole calculation on multi-objects contour extraction can be greatly decreased by the benefit of coarse-to-fine strategy and ideal noise resistance ability can also be expected in this algorithm.