Abstract: The log-polar transform is a well known space variant image registration scheme. However, due to the phase cross-power spectrum showing the multi-peak distribution, the accuracy of image registration is greatly influenced. To deal with this problem, an infrared image registration method based on the optimal peak of phase-correlation criterion was proposed. First, the source image and the reference image were converted to the log polar domain respectively. Second, the adaptive Butterworth high-pass filter was adopted to weaken the low-frequency noise. Finally, combined normalized cross-correlation and phase correlation similarity measure, the optimal peak of the cross-power spectrum was acquired based on truncating the low value of the cross-power spectrum. Moreover, the influence of LPT-image size on the matching performance was discussed and the appropriate size ratio of LPT transformation was presented. The experimental results indicate that the proposed method has higher precision and better anti-noisy performance than those of the normal LPT matching algorithm and is suitable for rotated, scaled and translated infrared images registration.