Repeated Median Filter with Adaptive Kernel Size

Y. Zaz*, K. Bouzouba, L. Radouane

LESSI, Département de Physique, Faculté des sciences B.P. 1797-30000 Fez, Morocco

* Corresponding author. E-mail:

Received : 28 April 2004; revised version accepted : 24 February 2005


Median filters are effective to remove sporadic noise from the image. We propose an Adaptive Repeated Median Filter (ARMF), it is based on the kernel size adaptation of Repeated Median Filter (RMF). This adaptation is computed according to the local intensity variance of each image pixel. It is shown that the ARMF is superior to several median-based filters in term of removing noise while simultaneously preserving sharpness (fine details).

In order to confirm our algorithm, the maximum entropy principle and the mean squared error (MSE) are used.

Keywords: Kernel size adaptation; Adaptive Repeated Median Filter; Local intensity variance; Maximum entropy principle.

© 2015