The behavior of two image registration techniques (the cross-correlation and the phase-correlation algorithms) for input signals corrupted by additive white noise is studied in the frequency domain. Useful expressions of the displacement measurement error variance are presented, which allows for the analysis of the accuracy of both methods. Both methods are more sensitive to noise for low-pass inputs (blurred images) than they are for high-pass inputs (textured, sharp images). The phase-correlation technique is more sensitive to noise than direct cross-correlation both for low-pass and high-pass inputs.
展开▼