Robust multi-frame image super-resolution using anisotropic diffusion

In the last three decades there have been many multi-frame image super-resolution (SR) algorithms proposed, but there are still many problems, such as edge blurring, artifacts are existing in the final reconstructed image, and many previous algorithms are sensitive to noise. This paper reviews some typical algorithms and addresses their limitations. In this paper, we propose super-resolution with the regularization based on anisotropic diffusion filter. We introduce new edge stopping function in our regularization process to hinder the outliers during estimation process in the flat regions and edge regions. We use guided image filter (GIF) to improve the smoothness in the final reconstructed image. The experimental results show that the proposed algorithm is robust to outliers and it can alleviate the blurness and artifacts problems in the final reconstructed SR image effectively. The Quantitative evaluation results show that our algorithm can achieve more average PSNR gain in the final reconstructed image as compared with other typical multi-frame image super-resolution algorithms.