Font Size: a A A

Structral Feature Matching In SAR Images

Posted on:2015-02-25Degree:DoctorType:Dissertation
Country:ChinaCandidate:T Z ChenFull Text:PDF
GTID:1108330509960953Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
Synthetic aperture radar(SAR) image is a very important remote sensing data in earth observation because it can be obtained in all weather at all times. With the further development of the platform and microwave imaging techniques, various types of SAR image data are acquired. In practice, SAR images often need to be aligned before they are compared. Image matching is to find the corresponding images or features of two or more images within the same scene, according to the physical and geometrical properties of the image. Hence, SAR image matching involves in numerous applications as the basic and common critical techniques.The structural feature is one of the most popular features used in SAR images matching. Due to the imaging mechanism of SAR and the complexity of the structural feature extraction, SAR images and their structural features have serious uncertainty, which seriously affect the performance of SAR image matching. Therefore, based on the analysis of the uncertainty in three types of typical structural feature, this thesis focuses on the robust, accurate and efficient structural feature matching method of SAR image. The principal contents and results are outlined as follows:Uncertainty analysis of SAR image and structural feature: A perspective overview of the imaging mechanism from the physical and geometrical characteristics of SAR image is reviewed. Then, the basic theory of the SAR image structural feature matching is introducesd. The composition and classficaiton of the uncertainty of the SAR image are presented and summed up exploratorily. The influences of the images matching method by the stochastic, fuzzy and incomplete uncertainties are analyzed. The characteristics of the SAR image matching are summed up. The feature selection criterion for feature matching of the SAR image is summarized. After that, the characteristics of three kinds of the structural features selected in the thesis are analyzed.Fuzzy contour feature matching of SAR images: A fuzzy contour feature matching method for blured SAR image is presented. The contour edge points extracted from a fuzzy SAR image locate imprecisely. It badly affects the performance of the feature matching. An invariant descriptor of fuzzy contour feature is realized based on the shape context descriptor and the Rough Set. The SAR image contour feature matching is implemented by the EMD measure. The performance of the method is verified by the modle matching and object classification experiments with field SAR images and the MSTAR data set. Results show that the method is applicable for the scene matching and object classification, the accuracy is improved.The uncertainty intersection feature matching of SAR images: First, a multi-level method based on the multi-scale theory and a fusion method based on the DS theory for linear feature extraction of SAR image are proposed respectively. They can all be used to extract linear features from the simple and complex background of SAR images. Then, an uncertainty model of intersection is presented for those linear features which are incontinuous and locate imprecisely. An uncertainty descriptor for the intersection is proposed based on fuzzy rough set. After that, a matching method for uncertainty lines is implemented using the intersection of line pair(or group) of the SAR image. The performance of the method is validated by image matching experiments of simple and complex scenario. Results show that the proposed method is suitable for the matching of those images which have obvious linear structural features, and the matching accuracy is improved.SAR point feature matching: Multiplicative speckle noise often significantly affects the accuracy and adaptability of the point feature matching method for SAR images. To address this problem, this study proposes a union matching method based on the SIFT and edge strength of the SAR image. Firstly, the initial SIFT matches are obtained by the normal SIFT operator, and the searching space is built. Using the geometry transformation model, the similarity is then determined. To get the optimal SIFT matches, the edge point matching based on pixel migration is applied. Furthermore, we outlined the optimization procedure of a union matching strategy. In the iterative process of global searching, the correctly matched tie-points are added individually. Finally, the accuracy, adaptability, and precision of the proposed method are validated through matching experiments on SAR images. Results showed that the proposed method is accurate and robust with respect to automatic matching of SAR images.Finally, all works in this thesis are summaried systematicly and the further researches are suggested.
Keywords/Search Tags:Synthetic Aperture Radar(SAR), image matching, structural feature, uncertainty, contour feature, fuzzy invariant descriptor, rough set, intersection feature, fuzzy rough set, pixel migration, Scale-invariant feature transform(SFIT), union matching
PDF Full Text Request
Related items