Study On Segmentation Of Remote Sensing Image Based On Level Set Method

The geometric active contour model based on level set method is well used for the segmentation of remote sensing images. However, the complexity computation of the model leads to a low efficiency. In order to satisfy the requirement of real-time, the algorithm efficiency need to be improved. In the other hand, remote sensing images are often severely polluted by various of noise, the robustness of the algorithm also need a further improvement.. Two classical geometric active contour model, CV (Chan-Vese) model and Regional Competition model, are taken into account in this paper, significant improvents are made in both efficiency and robust. The innovations are achieved as follows:Firstly, an improved numerical method for CV model was proposed. The method, by including a defined speed factor, achieves a better convergence rate, less dependence with the image gradient, so as to improve the robustness. Meanwhile, with the adjustment of the speed factor, the different objects number is obtained, to satisfy various situations and requirements, which improve the application of CV model.Secondly, a numerical method based on wavelet soft thresholding for region competition model was presented. The new method separates the problem of the energy functional of region competition model into two relatively simple ones. The solution of one simple function can be obtained by wavelet soft thresholding method, the solution of another function can be obtained by numerical calculation of known parameters. The time mostly consumes on the wavelet transform (WT) and inverse wavelet transform (IWT) of wavelet soft thresholding method, both of them can be solved by Mallat fast algorithm. Therefore, wavelet soft thresholding method can implement regional competition model faster.