Unsupervised Speckle Level Estimation of SAR Images Using Texture Analysis and AR Model

Abstract

In this paper, a new method is proposed for unsupervised speckle level estimation in synthetic aperture radar (SAR) images. It is assumed that fully developed speckle intensity has a Gamma distribution. Based on this assumption, estimation of the equivalent number of looks (ENL) is transformed into noise variance estimation in the logarithmic SAR image domain. In order to improve estimation accuracy, texture analysis is also applied to exclude areas where speckle is not fully developed (e.g., urban areas). Finally, the noise variance is estimated by a 2-dimensional autoregressive (AR) model. The effectiveness of the proposed method is verified with several SAR images from different SAR systems and simulated images. key words: AR model, equivalent number of looks (ENL), synthetic aperture radar (SAR), texture analysis.

A Fractional Lower-order Bi-spectrum Estimation Method Based on Autoregressive Model

“…If the excitation of a physical network by white noise ω(n) is viewed as a random signal of the power spectral density under normal conditions [14], then the p-order AR model can be established as:…”

Novel estimators of equivalent number of looks in polarimetric SAR imagery based on sub-matrices

“…(1) Unsupervised estimation The traditional ENL is estimated by selecting a region of interest, while some research has been done to develop a fully automatic estimation algorithm avoiding manual selection of a region of interest [17][18][19][20]. Since all these methods are related to only one polarization at a time, Anfinsen proposed a new unsupervised estimator based upon the polarimetric ML estimator [1] which follows the approach of Ref.…”