Remote sensing image registration, as the key technology in remote sensing image processing, is the crucial step of various operative technologies such as change detection, image fusion and image mosaic. In this paper we present a automatic remote sensing image registration approach based on SURF (Speed up Robust Features).We have done our research in the following fields:Firstly, we do some study in the area-based method and feature-based method in image registration. Then we compare several point feature extraction methods in aspects of speed, accuracy and repeatability. Because of the better performance of SURF feature, we decided to use SURF feature to extract the point feature in images.Next we introduce the concept of feature database. In order to further improve the efficiency of our automatic registration task, and reuse the feature we have extracted, we store those features into our feature database. The feature database makes it possible to fulfill our registration task between object image and database instead of between input image and reference image.Finally we demonstrate our automatic registration system. After analyzing the parameters and their influence on image registration performance, especially on speed and accuracy, we also present our approach to choose proper parameters. The experiment proves that SURF feature and its feature database has much practical value in remote sensing image registration. |