An example-based method in multi-frame super resolution

In this paper, a two-stage super resolution method that combines the multi-frame and example-based methods is proposed. Traditionally, the multi-frame super resolution method only utilizes the continuity prior and complementary information among the low-resolution(LR) images with sub-pixel misalignment. While the example-based method digs the prior from abundant training images, but performs low ability to process the severe blurring image. So in our paper, firstly, the sequence is processed by the traditional fast and robust super resolution method to enhance the definition. Then, in the second stage, the high-resolution feature (HRF)/high-resolution(HR) dictionary pairs is prepared. The near-high-resolution image acquired in the former stage is split into overlapped patches, then sparse coded to the HRF dictionary, and linear combined with the HR dictionary atoms. The experiments on the synthetic and real image sequence prove that the proposed method outstands from the other methods.