K-SVD Dictionary Learning and Image Reconstruction Based on Variance of Image Patches

The sparsity of signal is the premise of compressed sensing theory. It has been the hot topic for many years to sparsely represent the original signal accurately and quickly. For the sparse representation of image, the K-SVD dictionary training algorithm exhibits excellent performance. By calculating the variance of each block, different K-SVD parameters are settled, then the image sparse representation and Compressed Sensing reconstruction is achieved. Experimental results show that this method can preserve more image detail, and gain higher PSNR of the reconstruction results.