A fixed-point image denoising algorithm with automatic window selection

Abstract

A novel image denoising approach based on iterated median filtering is proposed. It is well suited for removing white noise and produces visually pleasing smooth surfaces while preserving edges and without producing artifacts. The denoised image is the fixed point of a nonlinear operator and can be obtained as the limit of a convergent sequence. We show that the sequence converges at an exponential rate. An algorithm implementing the proposed method is described; in addition to the fixed-point iteration, it automatically selects a suitable window and optimizes a scalar parameter based on an assumed noise level. The results of preliminary simulation experiments on well-known test images are presented. While the proposed method does not outperform earlier methods, it offers a theoretically well-understood foundation for future development; promising further research directions are discussed.