Landsat MSS和TM多光谱图像散粒噪声的去除方法

A method to remove shot noise in the Landsat MSS and TM multispectral images

针对Landsat MSS和TM多光谱图像上的散粒噪声问题,该文在分析其分布特点的基础上,提出了一种基于RX异常探测的噪声去除方法。视散粒噪声为光谱强异常,以RX算子探测异常,以光谱归一化方法定位散粒噪声。采用高同质化不规则的图像分割子区取代矩形窗计算RX异常,弱化小目标地物异常干扰采用不同噪声密度的MSS和TM多光谱图像开展算法验证。实验表明,此方法抑制了小目标地物异常,能有效去除MSS和TM多光谱图像上的散粒噪声,保留好像元,实现图像重建。

This paper proposes a method to remove shot noise in the multispectral images sensed by Landsat MSS and TM.In short,the RX operator is used to detected spectral anomaly,regarding shot noise as a strong spectral anomaly.A method of spectral normalization is also presented to position shot noise in the multispectral images.In order to weaken the small target anomaly,irregular high-homogeneous regions obtained by superpixels segmentation to a filtered image are introduced to substitute regular rectangular window for RX operator,experiments have been carried out on MSS and TM images with different shot noise density.The results both showed that small patch targets anomaly could be suppressed,the shot noise could be removed effectively,good pixels were unchanged and multispectral images were reconstructed.