Abstract: A multilevel median filter algorithm based on vertical and horizontal windows relation was proposed for the problem that traditional median filter algorithm can't protect the image detail very well, and that its performance will be in the sharp decline in dealing with high-density noise image. The algorithm determines the noise point of the image in window of NX N by using switching strategy. For the noise point, it's pixel value will be replaced by the median value of the median value in every window that lies in all 2N windows round the noise point. The pixel value will be retained for the non noise points. hherefore, we achieved noise removal and detail preservation. Experimental results indicate that the algorithm has better image detail preservation and strong denoising ability.

Key words: Vertical and horizontal windows relation, Salt and pepper noise, Multilevel median filter, Noise detection