Automatic Multi-object Segmentation by Two-phase Snake Processing
Cheng‐Hung Chuang, Wen‐Nung Lie · 2006
This paper presents a two-phase automatic snake algorithm for the segmentation of multiple objects from noisy or cluttered backgrounds. Traditional snake algorithms are often limited in their ability to process multiple objects and are required to have manually-drawn initial contours and fixed weighting parameters. Our algorithm features two phases: (1) the active-points phase and (2) the active-contours phase. In the first phase, grid points evenly distributed in the image are attracted and moved to form clusters near object boundaries. These clustered active points are then analyzed to obtain convex polygons as initial snake contours in the second phase, where a no-search movement scheme with space-varying weighting parameters is employed. Both the kinetics of active points and deformation of active contours accept our proposed adaptive gradient vector flow (AGVF) field as the contracting forces. Experiments show the stability of the AGVF field and good performance of our snake algorithm in segmenting multiple objects from noisy or cluttered backgrounds.