Application of Fourier Phase-only Correlation in Satellite Image Registration with Dem Shading Image

There is an optimization method for the geometric correction of Landsat images, which uses a simulated direct solar irradiance image made from a digital elevation model (DEM) and a solar direction to determine parameters of the geometric transfer equations. The optimization method is quite simple, and also allows us high accuracy geometric correction in sub-pixel order. However, promising in promotion of satellite image use for extraction of geographical information as well as for topographic correction, the optimization method still requires some improvements. In this research, we are focusing on the algorithm of optimization method for the positional difference of scene centers between given in the system information and in actual. In the conventional method, the offset of satellite image that originated in the positional difference of scene centers is calculated by two-dimensional cross-correlation analysis with the DEM shading image. The maximum value of the correlation coefficients is searched by a method based on a steepest descent method, thus the solution tends to hang on to local minima when the beginning value is not set appropriately. To solve the problem, we propose a new optimization method using the Fourier phase-only correlation (POC). The POC function requires no beginning value, and its maximum of the correlation coefficients is a sharp peak, thus it may be easy to find the positional difference of scene centers. In this paper, we have applied the proposed method to the system corrected Landsat TM images to investigate the features of the new method comparing to the conventional method.