The final step of our algorithm integrates the shift gradients across the image to construct a smooth shift map and render the target image. Since the shift for retargeting is usually one-dimensional, horizontal or vertical, the direct integration can still be inconsistent along the other dimension. This will again cause visual distortion. We propose an importance-weighted filtering method to address this issue. The method forces the integrated pixel shifts to be smooth along both dimensions and consistent across the image. The important pixels are weighted more so that the filtering pro- cess favors more on preserving their associated contents.