Sparse representations of images have revoked remarkable interest recently. The assumption that natural images admit a sparse decomposition over a redundant dictionary leads to efficient algorithm for image processing. In particular, the K-SVD method has been recently proposed and shown to perform very well for gray-scale and color image denoising task (\cite{elada},\cite{melada}). Meanwhile, the $TV-l^{\infty}$ model with special choice of dictionary has been proved to be very effective for image restoration(\cite{zeng.2006},\cite{lm2004}). In this paper, we propose a hybrid model which combines these two methods and may be regarded as a post-processing procedure for K-SVD. Due to the excellent work of K-SVD and the fact that $TV-l^{\infty}$ can reconstruct lost information quickly, this hybrid model lead to a new state-of-art denoising performance.