A new non-linear filtering algorithm with application to radar images

Under the assumption of log-normal speckle, the performances of various filtering algorithms (linear, homomorphic, α-linear, log-linear, generalized linear) are compared for processing radar images corrupted by speckle noise. The existence of a best estimator, the definition of which depends on the level of the speckle noise, is demonstrated. Theoretical investigations and case studies are discussed that show how to choose between linear filtering or log-linear filtering depending on the level of the noise in the image. If the noise level is high, the log-linear filtering is preferred; if the noise level is low, the linear filtering may be used