Interactive Image Segmentation Models And Algorithms

Image segmentation is one of the fundamental and key technology in image understanding and interpretation, whose aim is to divide the given images into non-over -lapping homogeneous regions that have their own characteristics. As an important branch of image analysis hierarchy, it is the focus and hot research in the image engineering field.Digital matting technology is a special image segmentation, and it has own an important dominate in a variety of applications now. It is a technology that separated interest foreground regions from its background. A small part of the image interest object and background regions is roughly scribbled, and according to these scribbles it can automatically and accurately receive the interest foreground object with the rules. Image processing techniques need to use a specific model to fit image feature data. This dissertation has studied several image segmentation methods based on different models.Firstly, a new texture image segmentation method based on adaptive window is proposed, and it can change the window size adaptively to extract better feature.Secondly, an image segmentation method based on the geodesic distance is discussed, which calculates the geodesic distance using the shortest path from each pixel to the scribbled pixels of the foreground and background. Obviously, it not only consideres the feature's difference but also the space information.Finally, the interactive image segmentation correction method based on geodesic active regions model is proposed, and it can correct the region and boundary probabilities using the correction algorithm base on scribbled pixels. Then it realizes the task of image segmentation and matting by minimizing the energy of the initial curver, and the segmentation result is improved in regional consistency and edge accuracy.