一种改进的RK-SVD随机噪声去噪方法

An Improved Denoising Algorithm Based on RK-SVD Method

摘要: 在图像的传递、形成和保存的过程中往往会受到外界因素的影响。稀疏表示是通过图像原子的信号组合来重现图像信号，而这些原子源自一个过完备冗余字典。字典的形成有两种方法，一种是设计字典来适应模型，另一种是使字典适应一组训练信号，以实现稀疏信号的表示。K-SVD算法是一种迭代方法，它在基于当前字典的列的稀疏编码和更新字典原子以更好地适应数据的过程中交替进行。本文在正则化K-SVD (即RK-SVD)算法基础上，通过改进了RK-SVD算法模型中计算误差项，使得改进的RK-SVD算法对数据的处理更加的准确，并且有效的阻止模型过拟合和欠拟合的发生。最后在实验的基础上，比较了改进后的RK-SVD算法的有效性。

Abstract: In the process of image transmission, formation and preservation, it is often affected by external factors. Sparse representation reproduces image signals by combining the signals of image atoms, which originate from an over complete redundant dictionary. There are two ways to form a dictionary, one is to design a dictionary to adapt to the model, the other is to make the dictionary adapt to a group of training signals to achieve sparse signal representation. K-SVD algorithm is an iterative method, which alternates between sparse coding of columns based on the current dictionary and updating dictionary atoms to better adapt to data. Based on the regularized K-SVD algorithm, this paper improves the calculation error term in the RK-SVD algorithm model, which makes the improved RK-SVD algorithm more accurate in data processing, and effectively prevents the occurrence of over fitting and under fitting of the model. Finally, on the basis of experiments, the effectiveness of the improved RK-SVD algorithm is compared.