Abstract
We propose a method to estimate image displacement by using RBF interpolation. This method can estimate image displacement such as parallel translation, rotation, and scale change at subpixel level. The evaluation function to be minimized is defined by the integral of square error between the interpolated images. Since RBF is expressed as the sum of Gaussian functions, we can compute the integral analytically and efficiently. We also present a method to reduce the computation time by efficiently selecting the region of summation.
