Decimated geometric filter for edge-preserving smoothing of non-white image noise

A procedure of recursive decimation is proposed to improve the performance of geometric filtering, in the case of spatially correlated image noise. Edges and fine textures are preserved, and noisy backgrounds carefully smoothed in a smaller number of iterations. Both SNR and subjective comparisons demonstrate an enhanced effectiveness with respect to geometric filtering. Such application fields as SAR speckle filtering and digital video processing are discussed and shown to benefit from the proposed scheme.