3D Rigid registration by cylindrical phase correlation method

Spatial alignment is an essential step before any further processing (such as fusion and change detection) of multiframe images can be done. We present a new algorithm that aligns translated and rotated pair of 3D images by means of phase correlation method (PCM). PCM is a computationally efficient method for translation estimation. We generalize a known polar-mapping approach of 2D image registration by PCM to estimate mutual rotation of a pair of 3D images about known axis. An improvement of this technique is given to eliminate influence of noise and image differences in non-ideal conditions. Finally, an iterative optimization procedure called cylindrical phase correlation method (CPCM) is proposed which uses these techniques in rigid body registration tasks. We utilize CPCM to register 3D tomographic images of human brain and study its performance in several experiments. CPCM shows extreme robustness to noise and is able to reliably and rapidly align even highly misregistered images.