Construction of vector field for snakes

Active contour models are important methods for image segmentation. Recently, active contour models based on vector field methods are focused. But vector field methods could not deal with long concavities efficiently. In this paper, a new vector field is constructed. In the view of different effects in active contour models, a vector field is constructed by two parts. A new conservative vector field is generated by an isotropic diffusion to maintain the contours´ basic shapes. In addition, a term from minimizing curl energy is used to generate the other vector field part to efficiently converge to the deep concavity. The active contour model based on presented universal vector field is not only insensitive to noise, but also fit for extracting concavity. Experiment results on synthetic images with long concavities and real images show the better performances of proposed method compared to the GVF and NGVF snakes.