The Study Of Large Scale Image Registration In Fourier Domain

Image registration is used to match two or more images of the same scene under different coordinates systems, which are taken under different photographic conditions. The technology is widely used in many fields such as computer vision, medical image processing and so on.Because of a small amount of calculation, robust and easy implementation, Fourier-based algorithms have got lots of attention. Traditional Fourier-based image registration technique used the discrete Fourier transform in Cartesian coordinates to approach the discrete Fourier transform in log-polar coordinate by interpolation. Although this method has a small amount of calculation, there will be big interpolation error. Based on this, the paper used the discrete Fourier transform witch is closer to log-polar coordinate to reduce the interpolation error.Main results of this thesis are as follows:(1) A new method of quasipolar Fourier transform approaching the discrete log-polar Fourier transform is presented. A quasipolar grid of equal angel is constructed. Thus, it needs interpolation in the radius direction of the quasipolar coordinate only. Experiment results show that our method is more accurate than the traditional Fourier-based image registration technique.(2) A new method of multilayer-peudopolar Fourier transform is put forward and a Multilayer-peudopolar grid is constructed encompassing pseudopolar grid and multilayer grid. This Multilayer-peudopolar grid is dense at low coefficients and sparse at high coefficients. Theoretic analysis and simulation results show that this grid is geometrically more similar to the log-polar grid than the existed methods. Experimental results show that Multilayer-peudopolar Fourier transform using the grid is more efficient for estimation of large scalings in images.