Level Set Based Image Segmentation Method

Image segmentation is a fundamental problem in almost all image processing and computer vision,and is very important for the success of image analysis. Recently, the level set method has received a great deal of attention especially for image segmentation. In the level set methodology, two-dimension (2D) level set can be visualized as the level surfaces of three-dimensional (3D) function, and a partial differential equation (PDE) controls the changes of the level surface. The level set method is appealing for its ability to handle topological changes automatically,therefor it is wildly used in image segmentation.This thesis concentrats on the level set methods and active contour model based image segmentation and presents some studies concentrated in the following two topics:1) Level set contour extraction method based on support value filter.This paper presents a new object contour extraction method, which combines the level set evolution with the support value filter. It analyzes image under the least squares support vector machine (LS-SVM) framework and uses the support values to represent salient features underlying image. The Gaussian filter, used in conventional level set method to compute the edge indicator, is replaced by the support value filter deduced from the mapped LS-SVM. The level set evolution method is implemented on the feature image obtained by convolving the support value filter with the original image. Experiments are undertaken on the synthetic and real images. The experimental results demonstrate that the support value filter can provide a good estimate for the optimal step size in a gradient descent algorithm and the proposed method has advantages over the direct level set evolution method in converging speed and contour extraction accuracy.The support value image was obtained by using support value transform. Then the level set method uses the support value image to compute the edge indicator. In order to avoid the effect of the eyebrow, the small parameter of edge indicator function first was used. After the evolving curve evolved across the eyebrow, the large parameter of edge indicator function is chosen. Then the evolving curve stops when it arrives to the facial contour. The experimental results demonstrate that the proposed method has advantages over the direct geometric active contours in converging speed and facial contour extraction accuracy2) Marker based level set method for image segmentationThis paper presents a novel image segmentation method based on the integration of level set and marker methods. The two distinct techniques for image processing are combined in a manner to utilize the strengths of both. The internal markers obtained by extended-maxima transform bring a priori knowledge to bear on the image segmentation. The initial level set function is constructed from region of interest (ROI) on the internal markers. In this way, an automatic initialization of the level set evolution can be obtained, and the boundaries of the objects can be extracted. The cost time does not depend on the size of the image but the region of internal marker because only level set function with markers is updated instead of the level set function for each pixel. Therefore, the consumed time is greatly reduced. The efficiency and accuracy of the method are demonstrated by the experiments on the real blood vessel images and MR images.This thesis researches the application domains of level set method and active contour in image segmentation, and improves the classical level set method.The new methods proposed in this thesis provide level set method and active contour model.