High accuracy subpixel image registration based on phase-only correlation of Harris strength
基于Harris角点量与相位相关的亚像素级图像配准方法

Phase correlation is used to perform image-to-image registration, and the fast direct Fourier-based algorithm(FDFA) makes registration accuracy up to subpixel, where Fourier components in the Fourier domain and windows in the pixel domain are carefully chosen to eliminate image-boundary effects, making the shift estimation more accurate. This paper proposed a new phase correlation method in which, instead of original images, used the Harris strength of them to determine the image shift. It first extracted feature points from the image using Harris detector, and selected the geometric point. Then placed a circular window around this point on its Harris strength, so the phase correlation could be carried out to estimate the shift. This method suppressed the image-boundary without introducing truncating error, and it did the phase fitting simply by the first Fourier component. Experimental results validate the accuracy of shift estimation is improved greatly by introducing the Harris strength for scenes which contain prominent geometric points.