中文摘要：

提出一种基于SVM分类与回归技术的图像去噪方法，即：去噪过程先利用SVM分类器将含噪图像中的像素分为噪声或非噪声点；接着，非噪声点像素值被保留，而噪声点像素值则通过SVM进行回归估计，从而达到去噪的目的，针对椒盐和高斯噪声在MATLAB6.5环境下搭建实验平台，运用OSU_SVM3.0和LSSVM工具箱分别建立4邻域、8邻域和24邻域3种分类和回归模型．去噪实验证明，与已有的算法比较，该方法能达到较高的峰值信噪比，具有很好的去噪效果．

英文摘要：

An image denoising method was hence presented by using SVM-based classification and regression technique. Firstly, the pixels in a noised image were divided into noise and non-noise pixels by using SVM classifier. Then the non-noise pixels were retained, and the noise pixels were one by one estimated with SVM regression model, so that the denoising was completed. An experimental platform was constructured in the environment of MATLAB6.5 for Salt ＆ Pepper and Gaussian noises, and three classifications and regression models with 4, 8 and 24 neighborhood domains were established by using OSU_SVM3.0 and LS_SVM kits. The experimental result showed that, by using this method and compared to the algorithms available, a higher PSNR could be achieved, showing a better donoising effect.