Small-Feature-Preserving Nonlinear Image Filtering in the Presence of Pulse and Multiplicative Noise

A new robust image-filtering algorithm, which preserves small features and efficiently rejects intense multiplicative noise, is presented. The filter uses M-estimates (robust maximum-likelihood estimates) and R-estimates (rank estimates, which follow from the theory of rank hypothesis-testing criteria). A two-stage filter is proposed, which provides a good visual quality of the processed images and provides a better performance in terms of the rms error and signal-to-noise ratio than standard filters.