Automatic snake contours for the segmentation of multiple objects
Cheng-Hung Chuang, Wen‐Nung Lie · 2002
This paper presents an automatic snake algorithm for the segmentation of multiple objects. Traditional snake algorithms are often short of the capability in processing multi-objects or required to have manually drawn initial contours. Our algorithm is composed of two phases: (1) the active point phase and (2) the active contour phase. In the first phase, grid points distributed everywhere in the image are moved to form clusters near object boundaries. These clustered active points are then processed to obtain polygons as initial snake contours in the second phase. Both the dynamics of active points and deformation of active contours are based on the gradient vector flow (GVF) field with a greedy search strategy. Experiments show good performance of our algorithm in segmenting multiple, concave, as well as overlapped objects, even in noisy images.