In this paper, we address the multiframe super resolution problem from a set of degraded, under-sampled, shifted and rotated low resolution images to obtain a high resolution image using the variational Bayesian methods. In the Bayesian framework a prior model on the high resolution image need to be specified, its aim is to summarize our knowledge of the image and to constraint the ill-posed image reconstruction problem. Appropriate prior model selection according to the super resolution scenario is a critical issue. Here we propose the one-parameter <i>l</i><sub>1</sub> prior. Experimental results demonstrate that the proposed method is very effective and compared favorably to state-of-the-art super resolution algorithms.