Block matching super-resolution parallel GPU implementation for computational imaging

This work presents the computational acceleration of a proprietary Super-Resolution (SR) algorithm (patented) for image and video enhancement. The considered algorithm is based on fusion SR techniques. The version proposed in this paper consists on the comparison of frames divided into Macro Blocks (MB) of fixed dimensions and on acquisition from a single camera. Due to its intensive computation, that limits its practical application in specific contexts where fast processing (even with real-time constraints) is necessary, the algorithm has been implemented in two platforms: OpenMP and GPU. Several tests have been conducted on seven popular image sequences and the results show a considerable improvement of the proposed solutions, in particular the Graphic Processing Units implementations. Consequently, it can be stated that GPUs represent an efficient solution to accelerate this type of algorithms to improve the perception of the image quality.