Object segmentation based on improved prior shape and CV model

Owing to great attitude variation of spatial objects and little grey difference between objects and spatial background, traditional Chan and Vese model can hardly get desired segmentation result. In order to solve problems that Chan and Vese model can’t segment correctly while some essential information is missed partly or some parts of the object are occluded, the integration of prior shape knowledge about the objects in the Chan and Vese model was given by Chan and Zhu. In this paper, a improved variational level set model with prior shape is presented to segment spatial objects under stars or cluttered earth background. The improved variational level set with prior shape constraint not only permits translation, scaling and rotation of the prior shape, but further introduces another two properties (shearing and different scaling of X and Y direction) in the energy functional model, which enhances prior shape’s self-adaption towards varying objects. Experimental results demonstrate that our model can achieve good segmentation towards spatial objects with great attitude variation in cluttered background.