A new method of multiscale image matching and application based on Fourier phase correlation

In order to realize high-precision registration of high-resolution remote sensing image, an improved multi-scale phase correlation matching method is proposed based on wavelet modulus value. This method takes the advantages of wavelet multi-scale antinoise characteristics by replacing gray-scale information in traditional phase correlation method with the wavelet module value, thus it not only improved the matching accuracy, but also had a strong robustness for the image with illumination changes, blurring and occlusion. Furthermore, a coarse to fine match is adopted to reduce the computational complexity according to the capabilities of wavelet multi-resolution analysis, by which the algorithm is initialized by matching two low-resolution images and increasing the accuracy by applying the coarse matching result to high-resolution images. It is possible to meet the real-time requirement. From the experimental results, automatic registration is quickly realized for high-resolution remote sensing image under the complex background condition with strong noise jamming. Comparing with traditional phase correlation method, the improvement on both the higher robustness on noise and the processing speed is obvious.