Group-based Sparse Coding with Adaptive Dictionary Learning for Image Denoising

Abstract

Group sparse coding (GSC) is a powerful mechanism that has achieved great success in many low-level vision tasks, showing great potential in image denoising. Traditional group sparse coding generally uses overcomplete dictionaries and $l_{1}$-norm to regularize sparse coefficients. But this is only an estimate of the solution, which cannot obtain a sparse solution and has a high computational cost. In this paper, we use a GSC framework with adaptive dictionary learning for image denoising. In order to improve the accuracy of obtaining sparse coefficients, the dictionary used in this paper is learned from the input image, which can be obtained by applying SVD once for each patch group. Then use ADMM algorithm to solve the objective function. Experimental results show that the PSNR value of our approach not only is competitive with many advanced image denoising methods but also achieves better visual effects.