Unsupervised estimation of speckle noise in radar images

Abstract

Abstract Speckle in radar images has the characteristic of a multiplicative noise. In this article, two unsupervised methods are introduced to estimate the speckle noise statistics using the mean and the standard deviation of small image blocks (4 × 4, or 6 × 6 pixels). Since most radar images contain many small homogeneous areas, a scatter plot of the standard deviation versus the mean can reveal the characteristic of the noise. The blocks from inhomogeneous areas have higher values for the standard deviation, and they are scattered above the main cluster. They are considered as outliers, and should be excluded in the statistical estimation. These two methods are designed for obtaining a linear fit in the scatter plot by ignoring outliers. Several synthetic aperture radar(SAR) images are used for illustration.