With the characteristic of all-time, all-feather and high-precision , the InSAR (Interometric Synthetic ApertureRadar) technology has been widely used to generate the DEM (Digital Elevation Model and detect the deformation of the earth’s surface. Removing the noise of the interference-stripes images is one important step during the data processing of InSAR. Most of he existed algorithms removed the noise with rectangle filtering windows without considering the anisotropic characteristic except the filtering algorithm based on the contoured filtering window. But it isn’t is to obtain the contoured window. Aim at the pattern’s anisotropic characteristic and the difficulty to obtain the contoured window , a method for InSAR interferogram filtering based on the center of interference-stripes is proposed. It first extracts the center of interference-stripes quickly by using CNN(Cellular neural networks) algorithm for it’s parallel feature to process data. According to the center lines the contoured filtering window is obtained. The InSAR interference phase images have jumping lines. If the filtering is just based on the images, the jumping lines would be shattered. So finally the sine or cosine filtering combined with the concourted filtering window is performed which can remove the noise of the InSAR phase images while avoiding damaging the phase distributing. The filtering of two fact interference phase images respectively from KASHI and SANXIA areas are studied. The experiment results show that this method is efficient.