Noise tolerance of power cepstra and phase correlation In Image registration
Dah Jye Lee, Sunanda Mitra, Thomas F. Krile · Annual Meeting Optical Society of America · 1988
Precise registration and subsequent matching of sequential images is a difficult task in many applications. Power cepstra of feature-based segmented images can be used for correcting the translational shift. The phase information in phase correlation technique is critical to the computation of the peak and is easily affected by the added noise. Since only the magnitude is involved in the computation of power cepstrum, the constraints that arise from the limitation of the phase components are absent in power cepstrum technique. The effects of the illumination change and the added noise are less than that on the phase correlation. Hence the power cepstrum registration method can have higher tolerance to noise. For typical feature-based images, the power cepstrum technique can detect and correct a spectral shift even when the SNR is as low as 0.17. The phase correlation technique for registration cannot be expected to have such high noise tolerance.