A fast deconvolution-based approach for single-image super-resolution with GPU acceleration

In this paper, we propose fast deconvolution-based image super-resolution (SR) with graphics processing unit (GPU)-accelerated computation. Recently, the deconvolution-based single-image SR has been proven to be very effective in upsampling images with favorable results. Based on the GPU-accelerated computation, we aim to realize the fast SR reconstruction and achieve balanceable performance in terms of both image quality and computational cost. To achieve this, we provide a novel and efficient deconvolution method to enhance the reconstruction results. We combine the gradient consistency in images with the anisotropic regularization which has been used in motion deblurring. Thus, we produce a directly parallelizable solution which is suitable for running on GPU by minimizing redundancy in computing. Experimental results demonstrate that the proposed method achieves superior performance in comparison with the existing methods with respect to image quality and runtime.