Remote Sensing Image Segmentation Model And Algorithm Study Based On Variational Level Set Method

Image segmentation, which is the particular application in remote sensing, is very important for image information extractions. Although there are hundreds of method for remote sensing images, no method is capable of segment all kinds of images. Remote sensing segmentation is far from automation. Up to now the most successful software of remote sensing image segmentation is the ecognition, whose characteristics is fractal net evolution algorithm and multiscale image segmentation. Even so, remote sensing image segmentation still front with several problems:(1) in most cases human have to intervene the image segmentation procedures;(2) in some cases though computer itself can automatically process image segmentation, we have to do much test in order to decided the parameters required by the algorithm programming in the computers;(3) low segmentation accuracy and unclosed boundaries etc, are still the problems bother us.In recent two decades, some new image segmentation techniques constantly emerging, and have been applied in remote sensing. In these new methods, the active contour model based on variational level set gained much attention. Active contour model alleviates some problems mentioned above. The most advantage of active contour model is that it has strong foundations of mathematics and can naturally deal with the topological changes of image boundaries. The energy functional in its model has clear physical and geometric meaning. The fact that the minimization of energy functional, i.e. the optimal segmentation, can be approximately obtained by variation theory makes it in recent been the hot issue in the field of image segmentation. From the former edge-based active contour model to now the region-based active contour model, the model of variational level set segmentation methods based on PDE's are always being improved and its application field have expended continuously.Traditional region-based active contour model has many drawbacks, such as time consuming, the energy functional is nonconvex which makes it easily stuck in local minima, etc. The objective of this article is to study the new algorithms to reduce the computation time, improve the segmentation result and enhance computation stability, especially study the region-based variational level set image segmentation method. This article combines the global convex optimal model, Split Bregman quick algorithm, which are all the new methods appearing in the field of variational level set, to perform the remote sensing image segmentation; expand the Split Bregman algorithm to multiphase region segmentation; devise different segmentation models for different images, such as pan, multispectral images and SAR images, discuss the selection of parameters and the applicability of these models. To addresses these problems, this thesis presented some studies concentrated in three topics:(1) The global convex segmentation model is been selected as the energy functional for two phase image segmentation. The Split Bregman quick algorithm is been used for speeding up the solution of this model. The novelty of this combination is:global convex segmentation model outbreaks the problems such as variational level set is easily stuck in local minima and been influenced by initial contour; the merit of Split Bregman algorithm is that it unites the Bregman iteration and split technique, realizes the quick segmentation, speeds up the computation greatly.(2) propose a new variational level set model integrating the edge detection and noise distribution statics for image segmentation, expand SB algorithm to multiphase image segmentation, deduce the iteration formula for this new model and corresponding SB algorithm.(3) Considering the characteristics of different types of remote sensing images——optical image and SAR image, select different edge detection function and noise distribution model respectively. The edge detection function and noise distribution model are substituted in the proposed new variational level set model. The corresponding solution flow is given.The thesis tests the proposed algorithm by remote sensing optical image, simulating SAR images, and true SAR images. The experiment results verified the correctness and effectiveness of the proposed algorithms.