Abstract: In this paper, a new edge-reserve filtering algorithm is presented. It adaptively adjusts the size of the filters supporting area according to the local variance of the interested area. The bigger the local variance is, which means the more high frequency texture information is contained in this area, the smaller the support area will be. In order to protect the edges of the image when it is filtered, this algorithm will divide the pixels in its supporting area into several groups according to the angle between its direction and the mean gradient of this area. By this way, pixels along the content edges will be given a bigger weight factors than pixels that are vertical to the image edge. Thus protect the edges from being blurred by the filtering process. The simulation results show that this algorithm is better than median filter both in terms of PSNR values and subjective image qualities. For pictures that are rich in texture information, the PSNR gain could reach 3dB higher.