Multimodal image registration based on SURF and KD tree

Abstract

Aiming at the requirements of multimodal medical image registration for good robustness, high-accuracy and speed, this paper proposes a registration algorithm of multimodal brain medical image based on Speeded-Up Robust Features (SURF) and K-dimension (KD) tree. This algorithm first of all extracts SURF feature points from images and creates feature vector, then build KD tree to complete the image matching, and finally the image registration process is accomplished by estimating space geometric varied parameters according to the matching point pair. The algorithm combines robustness of SURF and high efficiency of improved KD tree. Experimental results show that under the conditions of images with noise, non-uniform intensity and large range of the initial misalignment, the proposed algorithm achieves better robustness, higher speed as well as good registration accuracy.