Abstract: In the process of non-uniformity correction for the infrared images using the interframe registration based least-mean-square-error non-uniformity correction algorithm (IRLMS), the accurate estimation for the displacement of the moving foreground was lacked in the scenes where moving foreground exists and the the rate of calibration parameter can not be adaptively updated in control when the registration accuracy is low, thus ghosting would be produced. To solve this problem, an improved interframe registration based least-mean-square-error non-uniformity correction algorithm was presented in this paper. LK optical flow was used to re-registrate the moving foreground, estimating the motion displacement and correcting the error image, to overcome the ghosting caused by the moving foreground. Meanwhile, the peak of the phase correlation matrix of the images which has eliminated the non-uinformity can be estimated, and the peak can be used to correct the updating rate of parameters adaptively, to overcome the impact causing by the overquick updating rate of the calibration parameter. As is shown in the experimental results, this method can suppress the ghosting phenomenon in the case that the foreground is moving and the registration accuracy is low, and improve the practicability of the IRLMS algorithm effectively.