基于边缘剔除与空间去相关的图像信噪比评估方法研究  

SNR Estimation for Remote Sensing Images based on Edge Extraction and Spatial Dimension Decorrelation

作　　者：朱博[1] 王新鸿[1] 唐伶俐[1] 李传荣[1] 

机构地区：[1]中国科学院光电研究院,北京100094

出　　处：《遥感技术与应用》2012年第1期39-44,共6页Remote Sensing Technology and Application

基　　金：国家863计划项目(2008AA121805);广西科学研究与技术开发计划项目(桂科攻0993002-2)

摘　　要：分析地物光谱特征可知,同种地物具有相同或相似的光谱特征,在同一波段的光学遥感图像内就会具有相同或相似的灰度值。而且同一均匀区域内的同种地物信号之间具有很高的相关性,噪声则是随机的、独立的、不具有相关性信息的干扰成分。也就是说均匀区域内同一景物的光谱反射值是相关的,如果能估算出该点的信号值,就可以将信号与噪声分离。通过提出边缘剔除空间维度去相关法(EESDD)来估计图像信噪比,即首先剔除地物边缘,然后利用均匀区域同种地物的高相关性进行多元回归运算来拟合地物像素灰度值的方法估计噪声标准差,从而计算信噪比。该方法既可用于单波段图像,也可用于多光谱(含高光谱)图像的信噪比评估;既可用于图像中均匀区域也可用于不均匀区域的信噪比评估。通过与局部方差法(LMSD)以及基于边缘剔除的局部方差法(EE-LMSD),对几幅地物覆盖复杂程度不同的实测图像信噪比估算结果的比较得出,边缘剔除空间维度去相关法的结果要优于前述两种方法,而且更具健壮性。It is known that the same or even similar spectrum features and quantized data exist in the same type of the surface features in an image by analyzing the spectrum and surface features.And there are many correlations between signals in homogeneous regions.But the noises are random,independent and de-correlation.It means that the spectrum reflecting values in homogeneous regions of a remote sensing Image are correlating.The noises can be separated from the values,if the real signals can be estimated based on the correlation.The paper presents a method,called extracting edge and spatial dimension correlation（EESDD）,to estimate the standard deviation（SD） of signals or signal-to-noise ratio（SNR）.The method is full use of the surface features to extract the edges,and estimate the SD or SNR by the correlation of signals in homogeneous regions.EESDD can be applied in estimating noise not only for a single wavelength image,but also for a hyperspectral image.And this method is used for the homogeneous and inhomogeneous regions,too.Comparing to the local mean standard deviation（LMSD） and the edge-extracted local mean standard deviation（EE-LMSD） in Hyperion data with different scene contents,the result of EESDD is far better than those of LMSD and EE-LMSD.

关 键 词：噪声标准差 噪声评估 边缘剔除 空间去相关 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]