| During the process of image collecting, both the limitations of the ImageAcquisition System and the influence of the surroundings make it almost impossible toobtain all the information from the original scenario. How to dig deeply into theinformation included in image and to improve their spatial resolution have always beenthe problems we dedicated ourselves to in the image processing area. Imagesuper-resolution is believed to be a quick method to solve this problem.At present, image super-resolution has a lot of classic algorithms which achievegood results. However they have ignored the nonlocal similarity of image, resulting inthe waste of the image information. In this paper, on the basis of the context-basedinterpolation algorithm, we propose a practical nonlocal similarity-based single imageinterpolation algorithm. The basic idea is to regard the image interpolated by classicalgorithms as a noisy image, and the interpolation error as the image noise. So that wecan modify the interpolation result using the Non-Local Means denoising algorithm. Inaddition, we have applied the nonlocal similarity to the video sequencessuper-resolution reconstruction. In this case, in the super-resolution reconstruction wecan use the information from the nonlocal similarity patches not only in a single frameimage but also in neighboring frames. Experimental result shows that the edge of theinterpolation image achieved using the method in this paper is sharper than before, andthe artifacts have got a dramatical reduction. |
