Infrared And Visible Image Registration Algorithm

Image registration is an important part in image processing and computer vision. As the development of multi-spectral and multi-sensor imaging technology,image registration becomes more important. In fact,infrared image and visual images have their own properties and they have not very much similarity.So some traditional registration algorithms can not used for this situation,such as intensity correlation based,corner based,Fourier translation based method and so on. While,their edges and information entropy are similar.The algorithm based on edge is investigated in the dissertation. Two methods is usually used for edge matching, which are slow and not robust.So a new edge matching algorithm named edge normalized correlation method is proposed which is more quick and robust. Mutual information based method is also used for registration of infrared and visual images which have an evident defect that it costs too much time. So a new method that combines wavelet decomposition and Powell method is put forward which improves registration speed effectively. The simulation results show that both of the new algorithms are effective for infrared and visual images.