Achieving sub-pixel registration accuracy for radar imagery

This paper describes a robust and computationally efficient procedure to obtain sub-pixel registration accuracy. The new approach is based on the well-known phase correlation method which replaces image correlations with multiplications in the Fourier domain. Important to the new process is the observation that the phase correlation matrix has unity rank. This allows the phase shift information to be found in a rank one subspace with two one-dimensional reduced rank autoregressive weight vectors. The reduced rank subspace employed is not based on the common eigenvector approach but on reduced order correlation kernel estimation technique (ROCKET) that maximizes the mutual information criteria. The performance results show that the new algorithm is very robust to aliasing and edge distortion and can achieve high registration accuracy with reduced computational complexity compared to several recently published sub-pixel registration algorithms.