非子采样Contourlet变换系数统计建模及图像去噪应用.pdf

舨兴与 电子学进展 47，051005(2010) LaserOptoelectronicsProgress ~2010中国激光杂志社 doi：10．3788／lop47．051005 非子采样 Contourlet变换系数统计建模及 图像去噪应用 牛彦敏 王旭初2 lfz重庆重大庆学光师范电技大术学及信系息科统教学与育工部程重学点实院验，重室庆400030 1 ， 重庆400044J 摘要 融合拉普拉斯分布及广义高斯分布模型对非子采样Contourlet变换(YSCT)系数进行了统计建模分析。研究发 现，NSCT作为平移不变 Contourlet变换，系数在不同尺度和方向上均存在较大冗余，在广义高斯分布性等方面需 引入参数加 以约束。根据建立的统计模型进行了医学图像去噪实验。结果表明，和 Contourlet及NSCT软硬阈值去 噪等比较，该建模方法提高了噪声估计精度，增加了峰值信噪比，改善了图像视觉效果。 关键词 图像处理：Contourlet变换；高斯混合模型；图像去噪 中图分类号 0436 OCIS 100．2980100．4994 文献标识码 A StatisticalM odelingofNonsubsampledContourletTransform CoefficientsandItsApplicationtoImageDenoising NiuYanmin WangXuchu f。CollegeofInformationScienceandEngineering，ChongqingNormalUniversity，Chongqing400030，China1 l KeyLaboratoryof OptoelectronicTechnologiesandSystems，MinistryofEducation，Collegeof l 【 OptoelectronicEngineering，ChongqingUniversity，Chongqing400044，China J Abstract A Laplace and generalized Gaussian mixture distribution—based method is proposed to explore the nonsubsampledContourlettransform (NSCT)coefficients．TheinvestigatingresultrevealsthattheNSCT，asa shift—invariantcontourlettransform，obtainsredundantcoefficientsineachscaleandeachdirection，anditscoefficients differfrom thoseofContourlettransform inaspectsofgeneralGaussiandistribution．Theregularizedparametersshouldbe introduced to generalized Gaussian distribution modelto enhance the coefficientsdistribution．Themedicalimage denoising experimentalresultswith comparison to similarContourlet—based methodsindicate theproposed modeling methodimprovestheaccuracyofnoiseestimation，increasesthepeaksignal—noiseratio，andachievesbettervisualquality． Keywords imagep