The effect of quantization on SAR parameter estimation

The multiplicative model is commonly assumed for SAR statistical description, and it implies that the noise level is proportional to signal level. SAR digital imagery is usually available in linear detection with uniform quantization. This quantization can severely affect the estimation of statistical parameters for SAR data, mainly for low and high signal levels, because of the existence of a limited number of possible values. This may lead to an underestimation of scale parameters, like the standard deviation, and derived quantities (coefficient of variation -CV-, Li's variance ratio parameter etc.). In this paper images composed of segments of different average levels and textures are analyzed and areas where the estimated variance of the underlying clutter is negative are identified and carefully scrutinized through a Monte Carlo experience using the G/sub A//sup 0/ distribution.