A grey-level image alignment algorithm for registration of portal images and digitally reconstructed radiographs

An algorithm for automatic registration of pairs of portal images based on image correlation is presented. It uses a fast-Fourier-transform-based cross-correlation operator to find the optimal registration, accounting for both in-plane translations and rotations. Different cross-correlation operators have been tested: the Pearson linear correlation coefficient has been implemented by fast Fourier transform and its performance has been compared to that of the more conventional normalized cross-correlation. A sequential approach has been applied to speed up the registration considerably without degrading the performance of the algorithm. The algorithm has also been applied to the automatic registration of portal images to digitally reconstructed radiographs (DRRs), which have been modified to resemble megavoltage images. The results are indicative of the feasibility of this approach to the inspection of patient setup in radiation therapy. (C) 1996 American Association of Physicists in Medicine.