利用具有稀疏性、特征保持性和可分离性等特点的超完备字典的稀疏表示，基于核奇异值分解（K-SVD）算法，研究了对图像去除噪声效果以及影响效果的因素.理论分析及实验研究表明：K-SVD算法能够很好去除超声图像噪声，保留图像细节特征，获得更高的峰值信噪比（PSNR）值.在计算过程中发现K-SVD算法中的训练样本尺度大小是影响去噪效果的主要参数.

Using the sparse representation of over-completed dictionary which was sparsity,integrity and separability we studied the quality of image denoising and the factors which affected denoising quality in the framework of K-SVD algorithm. Theoretical analysis and experimental study show that the K-SVD algorithm can refrain noise very well, retain more details of the image and obtain better PSNR. The key factor which affects the quality of denoising in K-SVD algorithm is the size of trained pictures.