Contrast of automatic geometric registration algorithms for GF-1 remote sensing image

The geometrical registration of remote sensing image is an important premise for the subsequent processing of image. And it's also an important security for the application, such as agricultural condition monitoring. Different algorithms of automatic geometry registration lead to various registration effects. It's hard to meet the registration requirements of all images. Four testing types of plains, mountains, summer and winter were selected based the features of terrain and time. The main three registration methods were: cross correlation algorithm based on region gray, mutual information algorithm based on region gray and SIFT algorithm based on features. SIFT feature is the partial feature of the image, which can keep the invariance in rotating, scale-zooming and brightness changing. Then the automatic geometric registration was made for four classes of GF-1 remote sensing image using the above three algorithms. Two kinds of experiments were conducted for GF-1 remote sensing image under various conditions such as different terrains and different imaging time. The comparison of different geometric registration algorithms were made in the aspects of accuracy, efficiency and stability. The results show that the SIFT algorithm is the most appropriate one. The visual edge effect is good and the root mean square error reaches the magnitude of 10-5, which can satisfy the demand of precision. This method is simple and efficient, and it can be applied into agricultural condition monitoring and other business efficiently. © 2015, Chinese Society for Agricultural Machinery. All right reserved.