Abstract: Aiming at traditional methods of image denoising are unable to acquire define edge information of image.Then a new approach of image denoising based on support vector regression（SVR） is put forward. ε-SVR can achieve robust regression by introducing ε insensitive loss function.Regression is sparse and holds all the merits of SVM.In this paper theory of ε-SVM and its application in image denoising is analyzed.We compare image denoising effects between SVM and traditional image filtering.And we compare image denoising effects of SVM kernel function under the condition of different noises.Effects of order parameter of multinomial kernel on image filtering are also discussed.Experimental results demonstrate thatε-SVM works well for image denosing and has good performance of sparsity.

Key words: ε-SVR, ε insensitive loss function, image denoising

摘要： 针对现有传统的一些图像去噪方法难以获得清晰图像边缘的问题，提出了利用ε-SVR技术构建图像去噪滤波器的新方法。ε-支持向量回归机通过引入ε不敏感损失函数，可以实现具有较强鲁棒性的回归，而且回归估计是稀疏的，保留了SVM的所有优点。分析了ε-支持向量回归机理论算法及其在图像去噪中的应用，使用ε-支持向量回归机对图像进行滤波并且与最小值滤波、均值滤波和维纳滤波等常用的滤波方法相比较，还比较了SVM各种核函数对不同噪声的滤波效果和分析了不同阶数的Multinomial核的滤波效果。实验结果表明了ε-支持向量回归机能够有效地去除噪声，不但信噪比较高而且比较清晰，同时具有良好的稀疏性。

关键词: ε-支持向量回归机, ε不敏感损失函数, 图像去噪