A Deformable Image Registration Algorithm Using NURBS

Non-rigid image registration is an interesting and challenging research work in medical image processing, computer vision and remote sensing fields. In order to increase the registration accuracy, this paper designs a composite registration algorithm of a rigid transform and free form deformation. Because NURBS (Non-uniform Rational B Spline) with a non-uniform grid has a higher registration precision and a higher registration speed in comparison with B spline, this paper presents a free form deformation algorithm based on NURBS. On the basis of common similarity metric algorithms, this paper proposes a new MI (mutual information) arithmetic based on the matched region contour. We use snake model to extract the matched region contour in the contour-based MI algorithm. This algorithm combines gray level information of pixels with anatomical information. In consideration of the translation-invariant and rotation-invariant characters of steerable pyramid, we use steerable pyramid to decompose image in multi-scale geometric analysis. In our experiment we give the multiscale rigid translation results and the experimental results of the NURBS based FFD method quantitatively. The results show that this algorithm improves highly the registration precision.