PolSAR image denoising using directional diffusion
Romulus Terebeş, Monica Borda, Raul Malutan, Christian Germain, Lionel Bombrun, Ioana Ilea · 2016
We propose a novel directional diffusion method for denoising Polarimetric Synthetic Aperture Radar (PolSAR) images degraded by speckle noise. The method is developed using the Partial Differential Equations (PDE) formalism and it employs an edge detection function that takes as arguments the multiplicative gradient norm and locally adaptive thresholds. The diffusion process acts simultaneously on the amplitudes of the elements of the scattering matrix and is steered semi locally along common diffusion axis computed using a vector valued, structure tensor-based approach. The effectiveness of our approach in comparison with PDE and non-PDE state-of-the art methods for speckle noise removal, is demonstrated experimentally on computer generated and on real images, through visual comparisons and quantitative measurements.