Shearlet变换与核各向异性扩散的图像噪声抑制

吴一全^1,2,3, 叶志龙¹, 万红¹, 刚铁²

(1.南京航空航天大学 电子信息工程学院, 210016 南京; 2. 先进焊接与连接国家重点实验室(哈尔滨工业大学), 150001 哈尔滨; 3. 深圳市城市轨道交通重点实验室(深圳大学), 518060 深圳)



摘要:

为了更有效地抑制图像噪声,改善图像视觉效果,提出了一种基于非下采样Shearlet变换(nonsubsampled shearlet transform,NSST)与核各向异性扩散的图像噪声抑制方法．首先对含噪图像进行非下采样Shearlet变换;然后对所得到的低频和高频分量分别进行改进的全变差(improved total variation, ITV)扩散与核各向异性扩散(kernel anisotropic diffusion, KAD);最后对扩散后的低频和高频分量进行非下采样Shearlet逆变换得到噪声抑制后的图像．给出了实验结果,并且依据主观视觉效果和峰值信噪比、结构相似度两种定量评价指标,与近年来提出的基于小波阈值收缩结合全变差、基于复Contourlet域非线性扩散、自适应Shearlet域约束的全变差等3种噪声抑制方法进行了比较．实验结果表明,该方法的噪声抑制能力更强,且更为完整地保留了图像的边缘和细节信息．

关键词:  图像处理  噪声抑制  非下采样Shearlet变换  改进的全变差扩散  核各向异性扩散

DOI：10.11918/j.issn.0367-6234.2014.11.013

分类号:TN911.73

基金项目:国家自然科学基金(60872065); 先进焊接与连接国家重点实验室开放基金(AWPT-M04); 深圳市城市轨道交通重点实验室开放基金(SZCSGD201306);江苏省制浆造纸科学与技术重点实验室开放课题(201313); 纺织面料技术教育部重点实验室开放基金(P1111).



Noise suppression of image based on nonsubsampled shearlet transform and kernel anisotropic diffusion

WU Yiquan^1,2,3, YE Zhilong¹, WAN Hong¹, GANG Tie²

(1.College of Electronic and Information Engineering, Nanjing University of Aeronautics and Astronautics, 210016 Nanjing, China; 2.State Key Laboratory of Advanced Welding and Joining(Harbin Institute of Technology), 150001 Harbin, China; 3.Shenzhen Key Laboratory of Urban Rail Traffic(Shenzhen University), 518060 Shenzhen, China)

Abstract:

To suppress noise of image more efficiently and further improve image visual effects, a noise suppression method of image based on shearlet transform and kernel anisotropic diffusion is proposed. Firstly, a noisy image is decomposed by nonsubsampled shearlet transform(NSST). Then the obtained low-frequency component and high-frequency components are processed by improved total variation (ITV) diffusion and kernel anisotropic diffusion (KAD), respectively. Finally, the noise suppressed image is obtained by synthesizing diffused low-frequency component and high-frequency components through inverse nonsubsampled shearlet transform(INSST). Experimental results are given, in terms of subjective visual effect and two quantitative evaluation indicators such as peak signal to noise ratio (PSNR), structural similarity (SSIM), a comparison is made with three recent proposed noise suppression methods based on wavelet threshold shrinkage and TV, based on nonlinear diffusion in complex contourlet domain, and using TV with adaptive shearlet domain restraint. A large number of experimental results show that the proposed method has stronger noise suppression ability and preserves edge and detail information more completely.

Key words:  image processing  noise suppression  nonsubsampled shearlet transform  improved total variation diffusion  kernel anisotropic diffusion