An Efficient Distributed K-SVD Algorithm for Image Reconstruction

K-SVD, as a sparse representation method of the over-complete dictionary, has been widely used in image processing. However, the K-SVD dictionary requires to be bigger, resulting in a large amount of computation and longtime consumption during the experiment. Therefore, an efficient distributed K-SVD approach (DK-SVD+) for image reconstruction is proposed. Firstly, the image is decomposed into small overlapping patches, which can speed up processing and increase the precision of images that are reconstructed. Secondly, an optimized iterative shrinkage method is introduced in the sparse coding stage, which can speed up the convergence of the model. The experimental results illustrate that our proposed model performs better than the existing K-SVD and deep K-SVD models. At the noise level of 25, its average value of PSNR is 1.79dB higher than K-SVD and 0.15dB higher than DK-SVD, and it consumes the shortest time.