Researches On Registration Of Multi-source Remote Sensing Images Based On SIFT And Image Information

Multi-source remote sensing images processing can improve the effect of remote sensingapplication. Registration of the mutli-source remote sensing images is a key technology formulti-source remote sensing images processing. So the research on image registration formulti-source remote sensing images has great significance.This paper studies the registration for multi-source remote sensing images by combiningSIFT with some other image information. The main contents are as follows:(1) To make sure the research about remote sensing image registration work well, the firstpart introduces the background and significance of the research. And then to sum up and analyzethe relevant literatures, which provide an important theoretical basis for further study of theremote sensing image registration.(2) The second chapter briefly introduces the basis theories of image registration, includingthe definition of image registration, geometric transformation model, basic process of imageregistration and image registration evaluation criteria.(3) Taking into account the spectral information of remote sensing images, a new imageregistration method based on scale-invariant feature transform (SIFT) and vegetation indexanalysis is proposed. In this part, the traditional SIFT algorithm is firstly described in detail, andthen using the big difference of the ground spectral reflectance in multi-spectral remote sensingimages to introduce the vegetation analysis. In order to prove the validity of the new method, wemake simulation experiments on SIFT and the new method, and compare from the matching timeand match accuracy.(4) Taking into account of the gray information of images, a new image registration methodbased on SIFT and optimization of regional mutual information is proposed. The imageregistration scheme consists of two processes, namely the pre-registration process andfine-registration process. The pre-registration process is implemented by the SIFT approach witha reliable outlier removal procedure. By repeatedly fine turning several selected matched featurepoint coordination, a series of registration parameters are estimated by least square method andused to construct initial particle swarms. Next, the fine registration process is implemented toobtain the optimal match parameters by maximizing regional mutual information based on chaoticquantum particle optimization algorithm.(5) The last chapter summaries the whole work of the paper and put forward the future work on multi-source remote image registration.