Algorithms for subpixel registration

This paper presents an analysis of four algorithms which are able to register images with subpixel accuracy; these are correlation interpolation, intensity interpolation, differential method, and phase correlation. The subpixel registration problem is described in detail and the resampling process for subpixel registration is analyzed theoretically. It is shown that the main factors affecting registration accuracy are the interpolation function, sampling frequency, number of bits per pixel, and frequency content of the image. An iterative version of the intensity interpolation algorithm, which achieves maximum computational efficiency, is also presented. Analyses, computer simulations, and experiments for measuring displacements of objects using their speckle images have shown that this algorithm is faster than a direct intensity interpolation algorithm by a factor of more than ten thousand. Using bilinear interpolation and representing pixels by 8-bit samples, a 0.01 to 0.05 pixel registration accuracy can be achieved.