Abstract: Due to the major drawback of the millimeter-wave image having a fewer data and lower resolution, an improved self-snake model based on local variance is proposed for its smoothing.This method uses the local variance of window 3×3 to construct edge stopping function that is approximated to zero around fine structures and is close to one in the interior of homogeneous regions.Here the filtering process could be better control by constraining smoothing in regions with fine details while permitting effective smoothing in the interior of homogeneous.Comparative experiments certify that the proposed algorithm outperforms the corresponding conventional self-snake methods in terms of noise removal and feature preservation.