In this paper, we present a novel adaptive joint nonlocal means de-noising back projection super resolution method. The iteration of our SR method consists of two steps: ordinary back projection and the high-frequency data filtering for the reconstruction image by using joint nonlocal means denoising, in which self-similarity structures are obtained by the bicubic interpolation image. Furthermore, the content-adaptive filtering and fast self-similarity search strategy are very help for reduce computation complexity, while maintaining the performance. Despite our method conceptual simplicity, the experimental result show the proposed method can achieve comparable performance to other leading algorithms.