Abstract: To improve the performance of salt-and-pepper noise removal, a two-stage scheme is proposed. In the first phase, an improved adaptive median filter is used to identify pixels which are likely to be contaminated by noise (noise candidates). In the second phase, the image is restored using a variational method that applies only to those selected noise candidates. The proposed method can remove salt-and-pepper-noise with a noise level as high as 80%. Simulation results indicate that this algorithm is effective and better than traditional variational method based on adaptive median filter.