Removal of Interpolation Induced Artifacts in Similarity Surfaces

摘要

Registration of two images requires interpolation to generate a new image on a transformed grid, and the optimal transformation that maps an image to the other is found by maximizing a similarity measure. Similarity surfaces are subject to scalloping artifacts due to interpolation that give local maxima, and, in some cases, erroneous global maxima. We propose a new linear filter that is applied to input images and which removes scalloping artifacts from cross-correlation and mutual-information similarity surfaces. The computational burden is sufficiently low that it can be used in every iteration of an optimization process. In addition, this new filter generates image data with constant variance after linear interpolation, making measurements of signal change more reliable. Following filtering of MR images, similarity surfaces are smoothed with removal of local maxima and biased global maxima.