An Improved Orthogonal Matching Pursuit Algorithm For Fast Reconstruction Of Super Resolution Images

With the rapid development of information and communications technology, people are no longer satisfied with the text and voice as the single means of communication, but eager to enjoy the more vivid and intuitive visual experience via multimedia communications such as video and images. However, in the image acquisition process, due to interference of objective factors and down-sampling operation results from bandwidth limitations, the original high-resolution images lose many image details, which seriously deteriorates the quality of communication.Super-resolution technology can accurately restore high-resolution images by using single or multiple frames low-resolution images through the reconstruction algorithm. This technology can improve quality of image formation and maintain the hardware configuration unchanged. The compressed sensing theory is a novel signal processing theory. The theory can accurately reconstruct the original signal with signal acquisition at frequencies lower than the Nyquist sampling frequency. This paper aims to use compressed sensing theory to solve the issues related to super-resolution and finally improve image details.This paper introduces the concepts, mathematical models and application scenarios about the super-resolution and the compressive sensing. Then this paper proposes super-resolution reconstruction algorithm using the compressive sensing theory. The proposed algorithm applies compressed sensing reconstruction to super-resolution and achieves super-resolution via simultaneously estimating blur kernel while reconstructing the high-resolution image. In addition, this paper proposes an improved orthogonal matching pursuit algorithm. The improved algorithm significantly improve the efficiency of reconstruction via modifying the indexed atomic quantity at each iteration. Based on this, the proposed super-resolution reconstruction algorithm does not need any training set. Only a single frame image is used to achieve the reconstruction work.The paper implements the experiment to verify the feasibility and better performance of the proposed algorithm. Meanwhile, the paper compares and analyzes the reconstruction performance with different sparse basis.