Abstract: The contourlet transform is a new extension of wavelet transform，which provides a flexible multiresolution and multidirectional image representation.It can effectively capture the geometric structure of images.Because of the down sampling operation，contourlet transform is not translation invariant，which will introduces the pseudo-Gibbs phenomenon in the processing of image denoising.The cycle spinning and nonsubsampled are introduced to improve the performance of contourlet transform.The computational complexity，visual quality and dependence of the noise level are discussed based on the comparison between the two modified methods.

Key words: image denoising, contourlet transform, translation invariance, cycle spinning, nonsubsampled

摘要： 轮廓波变换是小波变换的新发展，是一种多方向、多尺度的几何分析工具，能更有效地捕捉图像的几何结构。传统的轮廓波变换因包含有下采样操作，使得它不具备平移不变性，从而会在图像去噪过程中引入伪吉布斯（Gibbs）现象。分别采用循环平移及非下采样滤波器组两种方法对其改进，着重从计算复杂度、去噪效果、噪声水平依赖性等方面对两种改进方法的去噪效果进行了对比。

关键词: 图像去噪, 轮廓波变换, 平移不变性, 循环平移, 非下采样