Phase correlation based only on the phase information is a highly accurate alignment technique which is insensitive to various types of noise and geometric distortion and immune to nonuniform illumination. But only pixel-level matching resolution can be obtained. This paper presents an efficient and practical image matching algorithm with sub-pixel accuracy. The algorithm is based upon phase correlation with rotational compensation and paraboloid surface fit and can determine the translational and rotational parameters needed to match the two images. The validity and accuracy of the algorithm are also discussed.
