An Optimal Algorithm for IR/Visual Image Registration Based on Main-Line-Pairs

An optimal algorithm for IR/visual image registration based on Main-Line-Pairs(MLPs)is presented.It is divided into four steps:we firstly extract all MLPs,which consist of two lines which satisfy certain conditions,from the based and sensed images,respectively;and then according to some criteria construct as much putative matching pairs between MLPs from the two images as possible to form a matching set;thirdly,exploit a branch-and-bound method to find a subset to maximize its total similarity measure and to minimize its registration error under the condition that each MLP from the two images at most belongs to only one element of the subset;finally,use all control points which are acquired from the optimal subset and affine transformation model to register the two images.The experiment results tested by IR/visual images are shown that the proposed algorithm for registration is accurate and effective.