QR Iterative Subspace Identification and Its Application in Image Denoising

Abstract

The foundation of compressed sensing (CS) is the sparse representation of signals. Over-complete dictionaries could be utilized to map signals into their sparse representation over the dictionary. And iterative subspace identification (ISI) is an effective algorithm to determine the over-complete dictionary from signal samples. In this paper, the QR decomposition is proposed to be employed in the ISI scheme so as to obtain the adaptive over-complete dictionary. It is shown that the QR-ISI outperforms the ISI in terms of the recovered PSNR. Finally, the QR-ISI method could be applied to image denoising. Experiment results are presented to show that the QR-ISI offers a feasible method for image denoising with reasonable performance.