Research Of Image Segmentation Based On Level Set And Fuzzy Clustering Methods

Image segmentation which is to extract the interesting object from its background in an image is a significant topic in image analysis for many years. It also is an important step in object detection and recognition. Image segmentation, which is playing important roles in remote sensing image segmentation, oil spill monitoring, product testing, fingerprint identification, optical character recognition, note identi-fication, radar image monitoring, medical image diagnosis, license information extraction, seal imprint verification, and watermarking security, has penetrated into different fields now.So far hundreds of image segmentation methods have been proposed at home and abroad. But no one kind of method can be applied to all image segmentations and achieves satisfactory segmentation result. The level set and the fuzzy clustering for image segmentation are mainly researched in this dissertation, the review focus on the recent studies of level set and fuzzy clustering, the work and the results obtained in this dissertation are as follows:The level set segmentation method is one of the famous image segmentation methods, the evolution of C-V level set function is also more complex, its main drawback is that the evolution is relatively slow, aiming at this problem, a narrow brand level set method based on region level set without re_initialization is proposed, this optimization method detect the approximate edge in low resolution image, then mapped the edge to the high resolution image, more accurate edges are detected in the narrow brand at the middle of the edge. A novel narrowband level set method is proposed, the gradient of the virtual distance function form a narrow band, where the active contour evolutes by simply adding and simply subtracting, the evolution has the following advantages:simple calculation, high-efficiency segmentation. This virtual distance narrowband level set method is also applied to LBF model, a novel level set model is proposed for the segmentation of the image with gray intensity inhomogeneity. Generally, if the grey means inside and outside the level set are almost equal, segmentation results are poor, so the C-V level set method based on level set and parzen-window probability density is proposed. Fuzzy clustering method is another one of the most famous methods for image segmentation without human intervention and the threshold, but the efficiency of image segmentation is low, aiming at this problem, a fast coastline detection method based on FCM is proposed. Meanwhile a novel fuzzy c-means image segmentation method is proposed, its effectiveness is due to two mechanisms, the first mechanism is the replacement of the Euclidean distance traditionally used to measure similarity of the image pixels by a novel similarity measure which is considered spatial neighborhoods using Gaussian kernel, and thus our method becomes less sensitive to the noise of the image. Another fuzzy clustering method with dynamic weights based on the kernel method is proposed, this method cluster data with noise features in high feature space mapped by the mercer kernel, not only can overcome the influence of noise feature vector on clustering, but also cluster the line and the non-group data without any experience.The target of the topic research is realized through combining theory and practice, some research is explored remote sensing image coastline extraction, oil spill segmentation, and medical image segmentation by using the level set and fuzzy clustering method. The coastline extraction of remote sensing image is very important for marine pollution monitoring, coastal region monitoring, coastline drawing, ship target, the spacecraft position and attitude controlling, remote sensing image registration, and cruise missile guidance. The oil spill segmentation of remote sensing image is the basis of oil spill identification, oil spill area computing, and oil spill drift velocity computing. The medical image segmentation is helpful to disease diagnosis for the doctor. And thus the topic research in this dissertation has its application value.