Skip to content

Active contours without edges

Why this mattered

Chan and Vese’s paper mattered because it shifted active contour segmentation away from the assumption that object boundaries must be found as strong image gradients. Classical snakes and geodesic active contours typically relied on edge indicators: they worked best when boundaries were sharp, continuous, and visible as intensity changes. The Chan–Vese model instead treated segmentation as a region-based variational problem, drawing from the Mumford–Shah functional and minimal partition ideas. In practical terms, a contour could converge to meaningful boundaries even when edges were weak, noisy, blurred, or partially absent, because the driving force came from fitting image regions rather than stopping at gradients.

This made active contours much more broadly usable. The level-set formulation allowed the curve to change topology naturally, so one initialization could split or merge and detect multiple objects or interior contours without explicit bookkeeping. The paper also made the method computationally concrete: it gave a finite-difference algorithm and demonstrated cases where gradient-based snakes failed. After this, “active contours without edges” became a standard template for region-based segmentation, especially in medical imaging, microscopy, remote sensing, and other domains where boundaries are often not clean edge features.

Its longer-term influence was not just a particular energy functional, but a change in what counted as a segmentation cue. Later variational, level-set, graph-cut, convex relaxation, and statistical segmentation methods built on the same idea that boundaries can be inferred from regional homogeneity, appearance models, or global energy minimization rather than local edge strength alone. Even as deep learning later displaced many hand-designed segmentation pipelines, the paper’s conceptual separation between boundary detection and region modeling remained central: modern segmentation systems still rely heavily on region-level evidence, shape/topology handling, and energy-like objectives that echo the shift Chan and Vese made explicit.

Abstract

We propose a new model for active contours to detect objects in a given image, based on techniques of curve evolution, Mumford-Shah (1989) functional for segmentation and level sets. Our model can detect objects whose boundaries are not necessarily defined by the gradient. We minimize an energy which can be seen as a particular case of the minimal partition problem. In the level set formulation, the problem becomes a "mean-curvature flow"-like evolving the active contour, which will stop on the desired boundary. However, the stopping term does not depend on the gradient of the image, as in the classical active contour models, but is instead related to a particular segmentation of the image. We give a numerical algorithm using finite differences. Finally, we present various experimental results and in particular some examples for which the classical snakes methods based on the gradient are not applicable. Also, the initial curve can be anywhere in the image, and interior contours are automatically detected.

Sources