Road Enhancement And Extraction In Remote Sensing Images Based On Nonsubsampled Contourlet Transform

Due to its good non-linear approximation performance for piecewise smooth functions in one dimension, wavelet analysis is widely used in signal processing field over the last decade. Wavelet is the optimal base for functions with point-like singularity, and its coefficients are sparse. Unfortunately, in high dimensional cases, wavelet analysis cannot take full advantages of the geometrical features that the data contained. It is not the optimal, or, the sparsest representation of the functions and easily damages anisotropic edges information in images. To solve this problem, the MGA (Multiscale Geometric Analysis) emerges as the times require, and its aim is to construct the most optimal representation for high dimensional functions.Nonsubsampled Contourlet Transform (NSCT) is a multiscale geometrical analysis tools, because NSCT coefficients are sparser but more accurate in linear target representation than wavelet coefficients, so NSCT is much effective in image denoising, fusion and enhancement. Therefore, the proposed algorithm is potential to achieve good performance either in segmentation accuracy or road localization. Moreover, NSCT has a fixed transform framework, which is convenient in the adjustment of the parameters of scale and direction.In this dissertation, NSCT based methods were discussed, and a new road extraction and enhancement method based on NSCT in remote sensing images is proposed, and integrate a variety of algorithms in one system.(1) A new road extraction method based on NSCT in remote sensing images is proposed, After the NSCT transform, feature extraction can be done and using FCM(Fuzzy C-means) clustering, followed by non-maxima suppression, preliminary results can be obtained, then after remove the false fragment and connect roads, then the final results can be get.(2) A new method of road enhancement based on NSCT is proposed. Image to be processed is first decomposed using NSCT into different scales and different directions, then road enhancement can be done using morphology in NSCT domain. And this is favorable to the road extraction.(3) A road extraction system based on the methods above is completed, this system including road enhancement, road extraction in remote sensing images, modify in alternating, this system can be used in road extraction in real remote sensing images.The work was supported by National 863 Program (No.2007AA12Z223) and the National Natural Science Foundation (No.60702062,60505010) for their support.