Remote Sensing Images Based On Wavelet Transform Automatic Registration Algorithm

Image registration is the process of matching two or more images that get from the same scene derived from different time, different sensors or different views of angle. Over the years, as the ultimate goal of image registration, automatic registration technology has widely researched. The amounts of remote sensing images will grow exponentially. To study fast and effective automatic registration of these images with high accuracies has become a critical problem in remote sensing.Based on the CBERS02's CCD images, we focus on the study of fast automatic registration of remote sensing images. The related work and main contributions in this paper are summarized as follows.Firstly, existing remote sensing images' registration algorithms and their merits and drawbacks are reviewed. By analyzing the present status, we found some problems about automatic registration. While the image becomes larger, the procedure of automatic registration becomes longer. Another, in the process of the automatic image registration based on point matching, a lot of false feature points are often extracted and real feature points are often excluded. Therefore, the key problems for the automatic registration focus on how to improve the registration speed and to improve the registration precision.Secondly, we study two methods of image registration: the methods based on grey-level information and the methods based on feature information. Further, we study two methods of automatic registration: the automatic local registration based on point matching and the automatic global registration based on grey-level information. Based on the experiments to the CBERS02's CCD images, we conclude that accurate feature points are very difficult to extract in the process of automatic registration. By comparing with these two kinds of methods, we select the first method for the CBERS02's CCD images' registration.Thirdly, based on the thorough investigation on automatic registration methods for remote sensing, a new automatic registration method is proposed by combining traditional registration algorithms with wavelet transform. In our method, an image pyramid includes a series of images with different size and different resolution is formed by using wavelet decomposition. An iteratively refining strategy is designed to reduce the search space and the time complexity. Theoretical analysis shows that our method greatly reduces computational complexity and improves the registration speed.Finally, an automatic global registration scheme based on wavelet decomposition and cross-correlation is designed. The procedure of this approach is fully automatic and computationally efficient. Compared with the automatic registration approach based on point matching, this method avoids registration errors caused by inaccurate and unevenly distributed feature points. Using wavelet decomposition, the approach reduces the search space and improves the registration speed.The experimental results demonstrate the efficiency and accuracy of the method. The automatic registration scheme is feasible to the CBERS02's CCD images.