基于尺度乘积和相关性的图像去噪

Image Denoising Based on Multi-scale Product and Dependencies

为了去除图像的噪声,提出了一种基于尺度乘积和尺度相关性的平稳小波变换图像去噪方法.在传统小波系数估计的基础上,考虑到尺度间的相关性,利用不同尺度小波系数形成的系数向量,通过线性最小均方误差估计小波系数,获得各个高频子带的估计系数.针对单纯利用尺度间相关性去噪造成的图像边缘失真问题,在不同尺度小波系数形成的系数向量中引入了小波系数乘积,不但可以较好区分边缘信息和噪声信息,而且提高了原有算法的去噪能力.仿真结果表明,该图像去噪算法能有效去除图像噪声,较好保持图像边缘,在峰值信噪比和视觉质量上都有较大提高.

In order to filter the noise in image, a method based on multi-scale product and dependencies using stationary wavelet transform for denoising is proposed. At the base of traditional coefficients estimation, the image is decomposited using stationary wavelet transform. And the dependencies of wavelet coefficients is considered. The wavelet coeffieients are estimated by the minimum mean square-error estimation using the wavelet coefficient vectors are formed with the different scales. Due to the problem of edge distortion using multi-scales dependencies in image denoising, the multi-scale product is added to the wavelet coefficient vectors. Then the edge information and noisy infornation are separated weU. And the denoiseing performance is improved. The experimental results show that the noisy image can be denoised well by the proposed algorithm. The edge information is preservated well. And the PSNR as well as the visual quality are well.