基于稀疏优化字典的图像去噪算法

Image denoising algorithm based on sparse optimization dictionary

稀疏表示算法是用过完备字典表示图像信息从而去除图像中的无用信息,达到去噪目的.KSVD字典是过完备字典中的一种,但是KSVD字典过于冗余,导致图像处理过程中冗余无用的图像信息降低算法的效率,为了提高KSVD字典的高效性和稀疏表示算法去噪能力,笔者提出了一种基于稀疏优化字典设计的图像去噪新算法.新算法的去噪步骤为首先运用正交匹配追踪算法求出稀疏系数;其次运用迭代算法用稀疏系数对初始DCT字典进行更新学习,在迭代的过程中逐渐去除噪声,得到去噪后的图像.仿真结果表明:与DCT字典算法、Global字典算法以及原有的KSVD字典算法进行对比,新方法的系数矩阵更加稀疏,去噪效果较好.

Sparse representation algorithm uses over complete dictionary to represent the image information to remove unwanted information in the image and achieve the purpose of denoising. KSVD dictionary is a kind of over complete dictionary, but the KSVD dictionary is too redundant, which reduces the efficiency because of the reduction and useless image information of the KSVD dictionary algorithm in image processing. In order to improve the efficiency of KSVD dictionary and the denoising ability of sparse representation, a new kind of denoising algorithm based on sparse optimization dictionary is proposed. In this algorithm, the orthogonal matching pursuit algorithm is used to calculate sparse coefficient firstly, then, the iterative algorithm with sparse coefficient is used to update learning initial DCT dictionary. The noise in the process of iteration is removed and the de-noised image is gained. The simulation results reveals that compared to sparse iterative algorithm of DCT, Global and KSVD dictionary, the coefficient matrix in this algorithm is more sparse and has higher signal-to-noise ratio （SNR）.