ROBUSTNESS OF WEIGHTED MEDIAN FILTERS BASED ON ESTIMATION APPROACHES
Rastislav Lukàč, Alexander Varga · Iranian journal of electrical and computer engineering · 2003
Weighted median (WM) filters, a nonlinear filter class based on a median operator and a weight vector associated with samples inside the filter window, take their popularity from the robust order-statistic theory, the noise attenuation capability and the degree of the freedom related to filter design. In order to adapt a filter behavior for a variety of statistics describing the desired signal and the noise distribution, there were developed some optimization algorithms based on estimation and structural approaches. In this paper, we focus on optimal weighted median filters based on the estimation approach. Besides well-known WM optimization algorithms that utilize linear and sigmoidal approximation of a sign function, we test and analyze a genetic approach that outperforms others optimal WM algorithms especially in terms of the signal-detail preservation.