As the characteristic feature sizes of integrated circuits continue to shrink, both research and manufacturing are increasingly reliant on the scanning electron microscope (SEM) images. In this article, we propose a robust algorithm that automatically extract,, the full edge contours from the SEM images and then converts the SEM images to pattern images even when the SEM images are corrupted by strong noise and lossy compression. This algorithm is expected to have many applications. An example of application-defect self-inspection based on SEM conversion using this algorithm, is also shown. (c) 2005 American Vacuum Society.