A robust and stable method for denoising images is proposed based on a novel sampling scheme and the recently developed generalized normalized gradient descent (GNGD) algorithm. The robustness of GNGD is suited for the proposed strategy which, using prior knowledge, resamples input data in order of decreasing (increasing) value before applying the adaptive algorithm. The approach facilitates a high level of denoising ability without causing the filtered data to experience unwanted distortion effects that afflict similar filtering algorithms. The performance of the approach is demonstrated by denoising simulations and comparison with other adaptive filter algorithms. Image data is used in simulations to demonstrate visually how distortion effects are kept to a minimum.