多级块匹配变换域滤波图像去噪

Multi-stage Block-Matching Transform Domain Filtering for Image Denoising

为了进一步提高非局部变换域滤波方法的图像去噪性能,提出一种多级块匹配变换域滤波方法.通过块匹配找到含噪图像中若干相似的图像块,然后执行相似图像块间的一维Haar小波变换,再用硬阈值收缩变换系数实现图像降噪;由于图像中充分相似的图像块数量是有限的,仅用一步上述操作并不能完全去除图像中的噪声,因此通过迭代策略去除剩余的噪声.实验结果表明,无论是PSNR值还是主观视觉质量,该方法的去噪结果都优于块匹配三维滤波方法.

In order to further improve the image denoising performance of the nonlocal transform domain filtering method, this paper proposes a multistage block matching transform domain filtering method. First, it finds some similar image blocks in the noisy image through the block matching, and then implements the 1D Haar wavelet transform among these similar image blocks, uses hard threshold shrinkage to transform coefficient for image denoising. Because the number of similar image blocks is limited in a single image, only one step above operations can not completely remove the noise in the image, so residual noise is removed by iteration strategy. Experimental results show that the denoising results of the proposed method are better than those of block matching 3D filtering method on both of PSNR and subjective visual quality.