Uncertainties in the estimation of ACF-based texture in synthetic aperture radar image data

Abstract

Spatial analysis of synthetic aperture radar (SAR) image data holds much promise for characterizing and discriminating environmental scene elements. The autocorrelation function (ACF) has been identified as a potentially useful spatial metric because it admits an analysis with conventional linear system theory. Recent models of spatial scattering suggest that ACF-based texture analysis of SAR image data is capable of discriminating between a variety of area extensive targets. The incorporation of texture in an image classification or segmentation system requires some understanding of the uncertainties in the texture estimates. In this paper, the authors introduce a particular ACF model and examine the errors associated with estimating its parameters from image measurements. They also conduct an analysis of two important classes of errors: imaging system errors and estimation errors. They found that as the proportion of raw signal used to create the image increases the effects of system errors rapidly degrade ACF performance. This has implications for operationally produced image products that do not use an autofocusing procedure. They also found that the agreement between theoretical and observed estimation errors is quite good, so that the scale of these errors may be accurately estimated during a spatial analysis of the image data. They found some residual bias that may be attributed to both the use of the ACF itself and to the way the ACF model was constructed.