The Research For Image Segmentation Based On Active Contour And Closed Form Matting

Image Segmentation is a process which subdivides an image into several nonoverlapping regions based on gray, color, texture and shape of an image. Image segmentation result has similar features in the same region, but has great differences between different regions. This thesis mainly studies segmentation method based on energy functional. This method refers to the active contour or some evolution model. The basic idea of active contour is to define an energy functional whose independent variables include contour curve, and then uses a continuous curve to represent the edge contour of the target object.The segmentation result of Chan-Vese model is very rough when dealing with the details of the target objects, which menas that it is unable to obtain the accurate segmentation result with complete details. Inspired by the closed form matting, we proposed a new model consisting of Chan-Vese model and closed form matting, keeping both advantages respectively. Then the variational method is utilized to solve the energy functional. Finally, the energy is minimized. After we get the minimum energy, the location of the curve is what we desired.However, techniques that attempt to segment regions using global region based model are usually not ideal for segmenting heterogeneous objects. As we consider the defects, there is a local region based framework which can be used to localize any region based energy. In this framework, the internal function needs to be modified, and then obtains a new energy functional. Finally, we present three internal energy measures inserted into the framework. Meanwhile, we reform the proposed model based on local region with the matting term. There are many advantages of modified local region based approach including robustness against initial curve placement and insensitivity to noise. The modified model not only can deal with heterogeneous image, but also incorporates the benefits of the proposed model.