Multi-temporal remote sensing image registration is the key step of change detection, and because of the remarkable difference and the probably unknown of sensor parameters, the automatic registration of different temporal remote sensing images is very difficult. Image registration based on Fourier-Mellin transform(FMT) is a global and phase correlation method, which is based on Fourier and Log-polar transform. This method finds the transformation parameters for registration of the images while working in the frequency-domain and it is resilient to noise, occlusions and so on. In this paper, an improved approach based on Fourier-Mellin algorithm is proposed for the registration. Spectrum aliasing and resampling interpolation will bring errors during Fourier-Mellin transform. To get a better registration result, we have improved it by adding window function and filtering to reduce spectrum aliasing and increase the robustness.
展开▼
