Abstract: To remove zigzag edges occurring in shift-linear interpolated image, this paper proposes an improved method based on the statistical features of the image. The basic idea is to estimate the edge pixels of the high-resolution image through the vectors and covariances obtained from its low-resolution counterpart according to their intrinsic "geometric duality". While nonedge pixels are got via shifted linear interpolation algorithm. No iterative computation is needed in this algorithm. The application of this new interpolation algorithm shows that it can keep the contours and details of non-edged areas.