An Improved K-SVD Algorithm and Its Application for Image Denoising

The KSVD algorithm has been widely used in image processing fields due to its high efficiency for image sparse representation. The traditional KSVD alternately updates the dictionary atoms and sparse coefficients through iterative operations. The algorithm is with noise suppression ability, and the reconstructed image can maintain the internal structure information of the processed image. Thinking about the intrinsic structure information can be acquired by the dictionary update procedure, in this paper, the principal component analysis (PCA) process is adopted to replaces the singular value decomposition (SVD) performing on the error term, and the first principal component is extracted as update item for dictionary atom. Furthermore, the improved KSVD is applied to image denoising. The simulation results show that the improved algorithm can better reconstruct the essential features such as edge and texture of the denoising image, and obtain higher peak signal-to-noise ratio while significantly reducing the time consuming of the algorithm.