GPU-ACCELERATED NON-LOCAL MEANS SUPER-RESOLUTION RECONSTRUCTION
Sun Kongzheng, Li Jiade, XU Shu-yan · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2013
Super-resolution is a process via fusing low-resolution image frames to obtain high-resolution images that typically requires highly accurate image registration.The modern non-local means super-resolution algorithm has successfully evaded the problem of image registration.However, its great computational impact has encouraged us to seek an approach that is able to make it more feasible for practical use.Thus in this paper, we propose an optimized highly efficient GPU-accelerated non-local means super-resolution algorithm.The proposed algorithm sufficiently utilizes the on-chip shared memory of GPU that ensures a significant performance boost.Experimental results show that the proposed algorithm not only overwhelmingly outperforms its CPU counterparts, but also achieves approximately four times performance increase than a typically conventional straightforward GPU implementation.