Abstract: Since traditional non-uniformity correction algorithm has slow convergence and inaccurate prediction, a multi-scale optical flow motion estimation based algorithm is developed. Random image noise between frames is filtered by time-domain low-pass filter, and the next frame is estimated by using the multiscale optical flow, then the non-uniformity correction is carried out. This algorithm is demonstrated in the thermal imager developed by us. Compared with the traditional non-uniformity correction algorithm, the convergence rate of this algorithm decreases form 60 frames to 25 frames; the precision of prediction is improved by 100%, and the corrected image variance is decreased from 76.539 to 32.482.