Spatially varying registration using Gaussian processes

In this paper we propose a new approach for spatially-varyingregistration using Gaussian process priors. The method is based on theidea of spectral tempering, i.e. the spectrum of the Gaussian processis modied depending on a user dened tempering function. The resultis a non-stationary Gaussian process, which induces dierent amountof smoothness in dierent areas. In contrast to most other schemes forspatially-varying registration, our approach does not require any changein the registration algorithm itself, but only aects the prior model.Thus we can obtain spatially-varying versions of any registration methodwhose deformation prior can be formulated in terms of a Gaussian process.This includes for example most spline-based models, but also statisticalshape or deformation models. We present results for the problemof atlas based skull-registration of cone beam CT images. These datasetsare dicult to register as they contain a large amount of noise aroundthe teeth. We show that with our method we can become robust againstnoise, but still obtain accurate correspondence where the data is clean.