The Research On Level Set Segmentation Method Based On Nonlinear Shape Prior Information

Image segmentation in digital image processing is to dividing the image into a plurality of regions with specific and unique property, and through these specific areas of interest to extract the target. The image segmentation is the basis of the computer vision and the techniques of the image processing, and provides the technical support for the standard of the multimedia data encoding. The situation of the study on the image segmentation will directly affect the level of research in these areas. So, there are more and more attention and concern on the image segmentation. Currently, the geometric active contour model, which is based on the theory of the level set, is the more popular image segmentation model. But, when the background of the image is too complicated or the target is obscured by other objects, if the traditional geometric active contour model is used, then the segmentation results are not ideal. So, it becomes an import way to introduce the priori information into the geometric active contour model, and it better improves the segmentation result by introducing the priori shape. In the algorithm based on Principal Component Analysis, it is to use linear form to reconstruct the samples when dealing with the Signed Distance Functions of the samples, but all of the SDFs of the samples are not represented via linear form, which means that most SDFs of the samples are nonlinear, and this will take the drawbacks and shortcomings to the formal segmentation, even make the final segmentation results not be satisfactory.In this paper, a novel segmentation model using the priori shape is proposed in order to solve the insufficient of the segmentation model using the traditional Principal Component Analysis(PCA) algorithm, and solve the intensity inhomogeneous human face images. The major works and innovations in this paper are:1. The Kernel Principal Component Analysis(KPCA) algorithm is adopted to reduce the dimension of the SDFs of the samples, then solves the corresponding eigenvalues and eigenvectors which are the basis of the proposed shape energy term.2. For MICO level set segmentation model is improved, the level set for each update, the update of the level set function when not used to establish the basis of its original image, while the use of each iteration of the level set in the same time, update its corrected image.3. Based on the theory of nuclear feature space, a shape energy term is built. The total energy term combines the image energy term which is the improved MICO segmentation model and the shape energy term which is the bettershape energy term proposed before.