Automatic registration of remote sensing image with moderate resolution

For automatic registration of remote sensing image, Image feature points extraction is the major difficulty because of its low level automation and slow computation speed. The effect is not ideal for applying point extraction algorithms of general pictures directly to remote sensing image feature point extraction. Focusing on the characteristics of remote sensing image, this paper puts forward corresponding improved methods. Firstly, the reference image is acquired automatically, then the image enhancement used histogram equalization is done for each band, at last, the image pyramids are built, and feature points are extracted band by band, layer by layer with the SIFT algorithm based on GPU. Results compared with Erdas Software show that the registration accuracy and speed are improved conspicuously.