Dual-scale medical image registration based on steerable wavelet

Image registration is indispensable to high-quality healthcare by integrating heterogeneous information in medical images captured from multiple imaging devices and at different imaging sessions. Multi-scale image registration has potential to achieve accelerated computation and to find global optimization through "coarse-to-fine" registration scheme. To retain these properties and meanwhile to avoid transformation error propagation in multi-scale registration, a dual-scale approach is proposed in this paper. Based on steerable wavelet with properties of translation- and rotation-invariance, the results of fast registration in the lowest resolution level are directly passed to initialize the registration in the highest resolution level. The experiments on medical images demonstrate that the proposed registration could be used for images with increased data volume.