基于奇异谱分析的高斯噪声降噪改进算法  

Gaussian noise de-noising optimal algorithm based on singular spectrum analysis

作　　者：李国芳[1] 王力[1] 龙飞[1] 

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《计算机工程与设计》2016年第8期2143-2150,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61263005;61065010);高等教育博士基金项目(20105201120003);中国新世界优秀人才基金项目(NCET-12-0657)

摘　　要：针对Donohue提出的多分辨分析小波降噪法中存在的恒定偏差、不连续性及重构图像失真等问题,引入奇异谱分析理论(SSA),对直接影响降噪效果的小波基、分解层数的选取和阈值函数进行改进。根据小波分解系数的奇异谱特性确定最优分解层数,通过小波降噪质量评价方法进行反复实验,对比分析选出最佳小波基,提出一种改进的阈值函数。仿真结果表明,针对加性高斯噪声人脸图像,该算法较其它算法能更好地保留有效图像细节信息,提高了算法实用性能,体现出更优越的数学特性和清晰的物理意义,减小了运算量。Aiming at the problems of constant deviation,non-continuity and distortion of reconstructed images in multi-analysis wavelet threshold de-nosing methods proposed by Donohue,improvements based on the singular spectrum analysis（SSA）were implemented on wavelet base,decomposition level selection and threshold function which directly influenced the noise reduction effect.The optimal wavelet decomposition level was determined based on singular spectrum characteristics of coefficients.An optimum wavelet base was selected by wavelet de-noising validity assessments and an improved threshold function was proposed with repeatedly and comparatively experimental analysis.Simulation results show that the improved algorithm can reserve more effective details to improve practical performance.Superior mathematical properties and clearer physical significance with less calculation were also reflected in additive Gaussian noise cases than other methods.

关 键 词：阈值萎缩 奇异谱分析(SSA) 最优分解层数 改进阈值函数 质量评价 

分 类 号：TN911.73[电子电信—通信与信息系统]