Research On Gray Image Segmentation Algorithms

People perceive over half of the information through vision. So, the development in the field of image processing does us a great favor to our perception of the outside. With the improvement of the processing abilities of the computers, and the maturity of the image processing techniques, digital images, which carry abundant information directly, become more and more important as the research objects in the field of computer science.In the research and the use of the images, people are only interested in some parts, we call them foregrounds. The rest called backgrounds. The techniques and the processes to divide the image into several parts which have different features and to pick up foreground are called image segmentation. On the other hand, the boundaries of the foregrounds are the edges we are interested in. Once these edges are detected, the foregrounds are picked up. These edge based image segmentation also called edge detection.In the beginning, this thesis will introduce some common edge detection methods and operators briefly, compare and analysis the different results processed by these operators. Then we will discuss the deformable models based segmentation. This method can be divided into two classes: parametric deformable model and geometric deformable model. In the first class, the boundary and the shape of the object are represented by the parameters explicitly, like Snake. In the second class, they are represented by higher dimension functions inexplicitly. We will focus on this model, give a particular description of these two classes in this thesis.Snake, also called parametric active contour, is an elastic curve defined in image field. The curve deformed driven by inner energy of the curve and outer energy correlated with the edges. The curve describes the boundary of the objects when the total energy reached the local minimum, so we can get a meaningful description of the boundary. We will expatiate on the theories, the algorithms and the instances. The models have some shortcomings So we will introduce corresponding improvements.The second class, also called Level Set Method, is built on a perfect foundation of mathematics. So it is good at dealing with the changes of the topologies, and always used in the segmentation of complex structures. We will focus on the Level Set Method; describe the theory, the algorithms and the important improvements about it. And I also made some improvements to the original Fast Marching Methods to process the point cloud. To enhance the performance and the efficiency of the level set method, we divide the pixels into two classes, then define different evolve speeds. The results of the experiments showed us the advantages of the new method.