Skip to Main Content
The problem of assessing the noise amplitude affecting remotely sensed hyperspectral images and the corresponding signal-to-noise ratio is discussed. An original algorithm for noise estimation, which performs the analysis of image bit-planes in order to assess their randomness, is described. Differently from more traditional signal-to-noise estimators, which need a homogeneous area in the concerned image to isolate noise contributions, this estimator is almost insensitive to scene texture, a circumstance that allows the developed method to carefully assess the noise amplitude of nearly any observed targets. The developed algorithm has been compared with the well-known noise estimator scatterplot method, for which a novel implementation based on the Hough transform is presented. Hyperspectral and multispectral data cubes collected by the following aerospace imagers, MIVIS, VIRS-200, and MOMS-2P on PRIRODA, have been utilized for investigating the performance of the two considered estimators. Outcomes from processing synthetic and natural images are presented and discussed along this paper.
Geoscience and Remote Sensing, IEEE Transactions on (Volume:45 , Issue: 8 )
Date of Publication: Aug. 2007