Texture feature based image registration

The aim of our research is to analyse the importance of texture information for registration of a DRR (Digital Reconstructed Radiograph) and EPI (Electronic Portal Image) medical images. In our research, texture features are extracted by Laws texture coefficients and used for computing registration criterion functions. The proposed feature based approach is compared to the commonly used approach, where a registration criterion function is computed directly from intensity features, i.e. grey values. For this purpose we observed accuracy of registration, the distinctiveness of local extrema and the distinctiveness of a global extremum of the criterion functions. These parameters are essential to achieve a correct image alignment. Our results show that for the given image modalities we can expect more robust and more correct registration when texture based criterion function instead of intensity based one is used.