Abstract:

An image segmentation method based on graph cuts and nonlinear statistical shape prior is proposed. Firstly, the input shape templates are registered in the input space, and the training sets are obtained. Secondly, the target shape prior is mapped to a feature space with principal component analysis by using a nonlinear kernel function, and the projected shape is obtained, which is mapped back to the original input space to obtain the average shape of the target, and thus forms a new energy function. Thirdly, through the weight coefficient selfadaptive adjustment of the shape prior term, the shape prior term of the energy function becomes adaptive to the image to be segmented. Finally, the image segmentation is accomplished by graph cuts technology so as to minimize the energy function. Experimental results show that the proposed method can not only correctly segment the images which are different than the shape prior templates, but also has better segmentation effect for the object images with occlusion and pollution.  Moreover, the proposed method can improve the quality of image segmentation.

Key words: kernel principle component analysis;average shape;probability map;image segmentation