Progressive image registration based on probability boosting tree

This paper proposed a novel progressive image registration method based on landmarks. At first, identification of corresponding control points is implemented by classifying of image features. We constructed a Probability Boosting Tree (PBT) to calculate the matching probability of each pair of corresponding control points and selected the highly matched ones as control points. Next, a progressive image registration is performed by inserting new landmarks selected by PBT one by one, which estimates the deformation model more and more accurate. Experimental results indicate that the proposed algorithm in this paper is feasible and can improve the precision of image registration significantly.