Using shape distributions as priors in a curve evolution framework

We propose a novel framework for constructing and using a shape prior in a curve evolution framework. The prior shape information is captured through shape distributions, which are histograms of features derived from the shape boundary. The resulting prior captures perceptual shape similarity, is robust to small sample size, and is flexible. We further derive a curve evolution force that corresponds to this prior. This enables us to use this prior to perform tasks such as mean shape calculation and image segmentation within a curve evolution framework.