A gamma filter for multi-look synthetic aperture radar images

This paper describes a new filter for multi-look SAR images, that is based upon their underlying Gamma distribution. The filter's first step is to estimate the shape and scale parameters of the best fit Gamma distribution from the local image intensity values. This is achieved by a maximum likelihood method that uses a numerical Newton-Raphson iterative technique. By adopting this approach no a priori expectation is placed on the number of looks used to form the multi-look image and thus a measure of the degree of correlation between each look is implicitly contained in the estimate of the shape parameter alpha. Variation in the underlying image intensity is captured by the estimate for the scale parameter beta, that depends on both a and the local mean. The filter's output is then given as the most probable value of the estimated Gamma distribution; this contrasts many extant techniques where the output is based on the mean. The new filter has been applied to simulated and 3-look Jers-1 images. Results show the filter accurate estimates the underlying Gamma distribution's parameters and reduces speckle noise. To improve performance at edges an adaptive implementation is proposed.