无抽样方向滤波器组用于图像去噪的方法研究

Research on image denoising using decimation-free directional filter banks

提出一种基于无抽样方向滤波器组的图像去噪新方法,首先,将一维半带滤波器(half band filter)转换成二维低通滤波器,通过对此滤波器的各种操作获得4方向、8方向和16方向等无抽样方向滤波器组,同时,将各频域方向滤波器转换成空域模板;其次,采用Contourlet变换中的多尺度分解方法,利用上述空域模板实现图像方向分解,获得噪声图像的各尺度多方向系数;最后,根据各方向系数的统计特性,合理设定去噪阈值,方向合成只需各方向子带图像相加,多尺度合成过程与Contourlet变换相同,完成图像去噪。实验结果表明:该方法不仅有效地去除了图像噪声,而且能很好地保留图像的边缘纹理信息,并很好地去除了Contourlet变换去噪中无法避免伪吉布斯现象所引起的视觉失真,与现有阈值去噪方法相比,图像信噪比明显提高。

This paper presents a new method of image denoising based on decimation-free directional filter banks.First,all the filters like four-direction,eight-direction and sixteen-direction directional filter banks are revealed by filters＇rotating,skewing and/or scaling.This paper transforms all directional filters in frequency domain to spatial operators,and uses multiscale analysis of Contourlet transform and spatial operators to analyze the input image and get coefficients.At last,the paper chooses the reasonable threshold based on statistic traits of all coefficients.Directional synthesis can be achieved by just simply adding all the subband images.Multiscale synthesis is the same as Contourlet transform.The experimental results indicate that the method avoids false Gibbs visual distortion and is better than the existing denoising algorithms in smoothing noises,preserving image textures and details and improving the SNR of image.