对椒盐噪声稳健的数字图像斯皮尔曼秩次相关法  

Robust digital imaging Spearmans Rho correlation for pepper and salt noise

作　　者：黄旺华[1] 王钦若[1] 徐维超[1] 胡忠[1] 周延周[1] 

机构地区：[1]广东工业大学自动化学院,广东广州510006

出　　处：《光学精密工程》2015年第6期1800-1806,共7页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.11072063);广州市科技计划资助项目(No.2014J4100203);广东省自然科学基金资助项目(No.1414050002003)

摘　　要：应用传统数字图像相关方法进行变形场分布测量时,很难精确处理带有椒盐噪声的数字图像。本文利用斯皮尔曼秩次(SR)相关系数对于椒盐噪声鲁棒的特点,提出了一种基于SR的数字图像相关算法。根据SR均值统计特征,当椒盐噪声密度足够小时,二元高斯混合模型数据的SR相关系数主要由散斑图像母体之间的相关系数决定。为了验证SR在椒盐脉冲噪声情况下的稳健性,采用可精确控制的模拟散斑图像进行了仿真,然后在三维扫描系统中使用SR相关技术计算出扫描数据的偏差。可控压缩散斑变形仿真表明,当椒盐噪声密度小于25%,计算子块大小为41pixel×41pixel时,SR数字图像相关仍可计算出精确的变形。将该算法应用于计算激光扫描纹理数据之间的偏差,结果显示,整体变形为4μm时,相关系数为0.996 6,表明SR相关系数具有较强的稳健性,在椒盐噪声下可用数字图像相关方法进行变形场分布测量。When traditional digital image correlation technique is used to measure the deformation displacement, the digital image with very small amount of pepper and salt noise is difficult to be processed precisely. Therefore, this paper proposes a new digital image correlation method based on Spearman＇s rho（SR） correlation coefficient robust to pepper and salt noise. According to the SR＇s average statistic characteristics, the SR coefficient of bivariate Gaussian mixture model is mainly decided by the clean speckle image when the density of pepper and salt noise is small enough. To verify the robustness of the SR to pepper and salt noise, the controllable simulation speckle image is used in a simulation, and the SR correlation technology is used to calculate the scanning errors. The results indicate that it can still get the accurate displacement vector by the SR digital image correlation when the density of pepper and salt noise is less than 25% and the block size is 41 pixel×41 pixel. The algorithm is used to calculate the laser-scanning texture data, and the sampling offset calculated is4μm with the correlation coefficient of 0. 996 6. correlation technique has strong robustness, and displacement vector under the pepper and salt noise. The result shows that the SR digital image is suitable for calculating the deformation

关 键 词：数字图像相关 椒盐噪声 斯皮尔曼秩次相关系数 变形场分布 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]