Blur invariant registration of rotated, scaled and shifted images

In this paper, we propose a blur invariant image registration method that can be used to register rotated, scaled and shifted images. The method is invariant to centrally symmetric blur including linear motion and out of focus blur. The method correlates log-polar sampled phase-only bispectrum slices, which are modified for blur invariance, to estimate rotation and scale parameters. Translation parameters are estimated using a blur invariant version of phase correlation. An additional advantage of using the phase-only bispectrum is the invariance to uniform illumination changes. We present also results of numerical experiments with comparisons to similar registration methods which do not possess blur in-variance properties. The results show that the image registration accuracy of our method is much better when images are blurred.