Automatic registration algorithm based on wavelet transform

Registration is a problem of estimating the displacement between image pairs. The conventional methods often cannot trade off between the estimation precision and computational complexity. This proposed method based on wavelet transforms is a multi-scale block matching one, which takes full advantage of the self-similarity that occurs after the wavelet transformation. This algorithm trades off between the estimation precision and computational complexity, because the precision is from coarse to fine, which reduces the search complexity of the motion vector. It has been proved that the convergence speed is exponential. This method takes full advantage of the precise location both temporally and frequently which characterizes the wavelet transform. Both the matching speed and consistency are boosted without the loss of matching performance.