Image Registration and Realignment using Evolutionary Algorithms with High Resolution 3D Models from Human Liver

K. Yan, L. Bertens, and F. Verbeek (The Netherlands)


3D realignment, CMA-ES, vasculature reconstructions,multi-object registration, Pearson correlation coefficient


This paper introduces a robust solution to the realignment of high-resolution image stacks as required in the study of micro-vasculature in human liver. We discuss the design and implementation of CMA-ES alignment algorithm that provides a better and more generic solution to the realignment problem compared to existing solutions. The result of a random rigid transformation test shows that CMA-ES clearly outperforms several known solutions on both accuracy and robustness. The CMA-ES yields an accuracy of approximately 80% whilst the hierarchical chamfer matching algorithm or the Marquardt-Levenberg matching algorithm have an accuracy yield that is less than 50 %.

Important Links:

Go Back