Research Of External Energy Term In Active Contour Model

Image segmentation and object boundary extraction is a very important branch of image processing and computer vision. It is the connection hub between low-level image processing and high-level image processing. Segmentation result directly affects subsequent image processing effect. Combining with partial differential equation and differential geometry, Active contour model(ACM) has become one of very important research subject in image segmentation. It has a wide range of applications in many areas. Therefore, the research of active contour model has both theoretical and application value.In this thesis, we first make a review of active contour model including present situation of development and basic theory. Then some classical vector fields of active contour model and shape prior based active contour model are detailed. Finally, we make a research of external energy term, and put forward some improved models in order to solve the problem of active contour model. The thesis focuses on external energy term and the main contributions are as follows:(1) The thesis expounds the basic theory of active contour model. From the perspective of energy functional, the energy functional of active contour model is boiled down into a basic form with two parts: internal energy term and external energy term. Besides,the research content are unified to external energy term of active contour model.(2) In order to solve the premature convergence problem caused by critical points of static force vector field, dynamic gradient vector flow model is proposed. The model establishes a relationship between vector field and contour. The vector field contains a dynamic force field and a static force field. The dynamic force field is used to push contour evolve to concavity and the static force field is employed to control the convergence of contour. Based on the different effects of these two parts, a new vector field is obtained by weighting these two parts. Compared with traditional vector fields, the constructed vector field improves segmentation result of images with concavity or complicated boundary.(3) A parameter shape prior model is proposed. Firstly, PCA is employed to train object shape. Transformation is added to shape representation. Secondly, through a combination of shape energy, ACM energy and shape constraint energy of kernel density function, object boundary extraction process is converted into parameters solving process of object shape. Aimed at the flaws of the traditional shape energy, two new shape energy functions are defined when objects are occluded by other objects or some parts of them are missing. The new shape energy functions have better segmentation result. Finally, a two-phase alternating iteration solving scheme is proposed for numerical, and reduces the complexity of the numerical solution.