Subpixel Shift Estimation Method for the Registration of Noisy Images Using a Wiener Filtered Local Region

we propose an accurate subpixel shift estimation method for the registration of noisy images. The proposed method uses a specific local region to perform phase correlation instead of the whole noisy image and applies an adaptive Wiener filter to a local region for suppressing noises. Firstly, a local region is selected in the vicinity of the feature points from Harris corner detection by analyzing gradient and distribution of frequency component. Next, adaptive Wiener filter which is varying the size of window depending on the property of a local region is applied to eliminate noises. And then, iterative phase correlation is performed in a restricted iteration range with coarse to fine method. In the experiments, the proposed method shows higher accuracy in registering noisy images than other methods.