Medical Image Registration Based on BSP and Quad-Tree Partitioning

This paper presents a study of image simplification techniques as a first stage to define a multiresolution registration framework. We propose here a new approach for image registration based on the partitioning of the source images in binary-space (BSP) and quad-tree structures. These partitioned images have been obtained with a maximum mutual information gain algorithm. Multimodal registration experiments with downsampled, BSP and quadtree partitioned images show an outstanding accuracy and robustness by using BSP images, since the grid effects are drastically reduced. The obtained results indicate that BSP partitioning can provide a suitable framework for multiresolution registration.