An Iterative Image Registration Technique Using a Scale-Space Model

Registration between two images is a key problem in computer vision. Current methods tend to separate the scale estimation process from translation and rotation estimation. This is due to the fact that the scale parameter is inherently related to the image resolution. In this paper, we present an area-based image registration technique that can simultaneously estimate translation, rotation, and scale parameters and take into account differences in resolution between two images. We first develop a scale-space model that relates the entire reference image pixels to a single observed image pixel with a scale parameter. This model is then easily generalized to include x-y translation and rotation parameters. By embedding this scale-space model into a non-linear least squares method, we can iteratively estimate the four registration parameters (x-y shift, rotation, and scale) in a unied manner. We test the validity of the proposed method on both simulated and real image data.