Feature Constrained Non-rigid Image Registration

In this paper, we present a novel hybrid algorithm for non–rigid image registration that integrates intensity–based and feature–based methods. An important feature of our method is that the associated optimization problem takes into account the intensity sim-ilarity, feature correspondence and smoothness constraints simultaneously. Hence, the resulting registration algorithm yields the appropriate transform that achieves good lo-cal feature mapping and global intensity alignment. We demonstrate the application of the algorithm using two different types of features: the landmark and outer surface. However, these two features are treated differently in the algorithm. The landmark cor-respondences are expressed as interpolation constraints in the optimization problem, so that the landmark pairs in the fixed and the float images are guaranteed to be perfectly matched to each other. For the surface features, we propose a ”regularizer ” based on the level–set shape representation into the optimization problem. Thus the resulting al-gorithm is able to enhance the alignment of surface features. We solve the optimization problem in the variational framework, in which the extended Euler–Lagrange equation characterizes the solution. The improvement of the developed algorithm is not limited by image dimension and image modality. Finally, we demonstrate the effectiveness of the proposed algorithm by experimental evaluation using monomodal medical image data.