A level set with shape priors using moment-based alignment and locality preserving projections

A novel level set method (LSM) with shape priors is proposed to implement a shape-driven image segmentation. By using image moments, we deprive the shape priors of position, scale and angle information, consequently obtain the aligned shape priors. Considering that the shape priors sparsely distribute into the observation space, we utilize the locality preserving projections (LPP) to map them into a low dimensional subspace in which the probability distribution is predicted by using kernel density estimation. Finally, a new energy functional with shape priors is developed by combining the negative log-probability of shape priors with other data-driven energy items. We assess the proposed LSM on the synthetic, medical and natural images. The experimental results show that it is superior to the pure data-driven LSMs and the representative LSM with shape priors. © 2013 Springer-Verlag Berlin Heidelberg