摘要
提出了一种基于非采样Contourlet变换与Wiener滤波相结合的图像去噪算法。非抽样Contourlet变换可以实现多分辨、局部、尤其是多方向图像表示,且具有平移不变性,避免了去噪过程中出现的伪Gibbs现象;而Wiener滤波在最小均方误差意义上是对图像的最优估计。该算法结合Contourlet变换和Wiener滤波各自的优点,首先采用非采样Contourlet变换对含噪图像进行多尺度、多方向分解,得到低频子带系数和各带通方向子带系数;然后在各子带图像上进行Wiener滤波;最后经过非采样Contourlet逆变换得到去噪估计图像。该方法应用于peppers图片去噪,结果图像峰值信噪比(PSNR)增加为15.6775,最小均方误差(MSE)减小为1749,好于Wiener去噪(PSNR为13.7618,MSE为2549)和小波与Wiener滤波相结合去噪(PSNR为14.0662,MSE为2734)。实验结果表明此方法提高了图像的峰值信噪比、减小了均方误差。
This paper proposes an image denoising algorithm based on nonsubsampled contourlet transform (NSCT) and Wiener filtering. NSCT can result in a flexible multiresolution, local, and directional image expansion. NSCT is shift-invariant, and avoids pseudo-Gibbs phenomena around singularities in image de- noising. Besides Wiener filter is the optimal estimation of image from the meaning of MSE. Firstly, the NSCT was performed on the source image at different scales and directions, thus the low frequency subba- nd coefficients and varieties of directional bandpass subband coefficients were obtained. Then, the low frequency and directional bandpass subband coefficients were processed by Wiener filtering. Finally, a denoised image was obtained by performing the inverse NSCT. The method was applied in peppers image denoising. As a result, PSNR was 15. 6775 and MSE was 1749, better than the results obtained by other methods such as direct Wiener filtering (PSNR was 13. 7618, MSE was 2549), wavelet transform com- bined with Wiener filter ( PSNR was 14. 0662, MSE was 2734). The experimental results deminstrate that our algorithm improves the PSNR of image and decreases MSE of image.
出处
《中国体视学与图像分析》
2009年第2期147-151,共5页
Chinese Journal of Stereology and Image Analysis
基金
中国博士后基金项目(20060401012)
青年学术骨干项目(01140303)