Subpixel Image Registration Using Cell Areas

Most image registration algorithms assume that the images to be registered are sourced from a single camera and captured with relatively short time in between while the camera or the scene is doing a simple and definable movement on the image plane. It is also assumed that images are generated or can be generated from a higher resolution (or continuous) image by appliying some affine operations, either linear or nonlinear. Under the same constraints and assumed relationships between image pixels and CCD cell areas, a simple subpixel registration technique for translation and rotation is proposed. We tested the new technique on artificially shifted images and real images and compared the results with the ones obtained using popular phase correlation method and concluding that our results is superior.