Abstract. We present a multi-scale approach for non-rigid image reg- 

istration of tubular tree-like structures such as vessels. Therefore, we 

consider a Gaussian and a combined morphological and Gaussian scale- 

space. Similar to a multi-level framework, we solve a sequence of registra- 

tion problems on several scale-space levels using a standard variational 

approach for non-liner volumetric image registration. The overall idea is 

to avoid locally ambiguous mappings between parts of the images by re- 

moving morphological details but also finding a global optimal solution 

by spreading remaining local information using Gaussian scaling. We 

successfully tested our method on registration of 3D CT and ultrasound 

images of hepatic vessels.