Subpixel high accuracy image registration for radar interferometry processes

Image registration is one of the important steps in processing remote sensing data. In the process of SAR interferometry (InSAR) there is use of multiple data sets from different times or different look angles. Generally, data registration processes contain two steps; the first step is estimation of the transformation of coordinates from the reference image to all other images in the data set. The second step is resampling (interpolation) of the images to the reference grid. This work presents a new algorithm for the estimation of local shift in SAR detected images. The algorithm uses a Fast Fourier transform (FFT) and a cross power spectrum of the two images. From the cross power spectrum parameters and some empiric fix parameters are calculated for the translation vector up to a fraction of a pixel. The proposed algorithm is invariant to the image gain (or uncelebrated images and change in dynamic ranges) and also to bias change, another benefit from the local shift estimation is the low influence by small area of image corruption. It also shows that our method is minimally influenced by noise and clutter that are inherent in detected SAR images. In this work we present both simulated and real data results and compare the algorithm to other methods of registration.