Remote Sensing Image Fast Registration Algorithm Combined With Wavelet Transform And SIFT Algorithm

Image registration is a fundamental problem in the field of remote sensingimage processing, in the realization of image fusion, image change detection,image segmentation, image correction, image mosaic and other applications, theimage registration is an indispensable key step. Due to remote sensing image dat-a is too large, resulting in image registration algorithm computationally intentive,registration takes too much time, and can’t meet the requirements of practicalapplication. How to carry out the registration of these images more effectivelyand develope fast, automatic, high-precision remote sensing image registrationalgorithm is in urgent need of sloving.In the field of image registration, SIFT(Scale Invariant Feature Transform)algorithm can achieve precise image registration when the scale, rotation,illumination, viewing angle, affine changes and other case exists between theimages to be matched, so it has been widely applied in the field of imageregistration. However, in the remote sensing image registration application,because the image data is too large, calculation for SIFT feature extraction andmatching is too large. It will result in more registration time and less efficiency.To slove above problem, Remote sensing image fast registration algorithmcombined with wavelet transform and SIFT algorithm is proposed and acoarse-to-fine registration idea is used to achieve image precise registration.Firstly, image are decomposed into multi-scale images using wavelet tansformand wavelet image pyramid is got. Secondly, the low frequency sub-images atcertain level of wavelet image pyramid are input to SIFT algorithm to completecoarse image registration. It makes the calculations of image registration isgreatly reduced and the efficiency of image registration is greatly improved.Finally, in order to improve the image registration accuracy, phase correlationmethod is used to match the coarse registration images and sub-pixel imageregistration is got. The registration results show, on the condition of one layerimage wavelet decomposition, remote sensing image registration algorithmcombined with wavelet transform and SIFT not only can greatly improves thespeed of the remote sensing image registration but also produces good imageregistration effect.