Statistical Deformation Model For Intensity Based Image Registration

The main goal of intensity based image registration is to find the spatial relation between images to be aligned without calculating corresponding salient features. In this article, we propose a new framework that incorporates prior registration examples in order to obtain smooth deformation maps. During the training phase, an elastic image registration procedure is used to obtain the deformation fields by modeling them as locally affine and globally smooth. Next, the estimated geometric transformation warps are used to obtain a prior deformation model. Based on second order statistics, we have used principal component analysis (PCA) in the steerable wavelet domain in to generate a set of orthogonal deformation bases. A smooth deformation is now guaranteed by projecting the locally computed geometric transformations onto the subspace of plausible deformations. The new algorithm was validated using the Amsterdam library of images (ALOI). The advantages of using the steerable wavelet analysis in principal component analysis are presented.