Abstract: This paper proposes an new denosing method based on fuzzy neural network of geometrical features using crack its own characteristics.Its basic idea is to firstly measure 8kinds of geometric feature vector for noise estimation as input to establish a fuzzy neural network system (FNN) to train many images.Then it use the trained FNN to judge whether the image points for the noise point,if the result is noise,it is denoised using the classical median filter,Otherwise,retained the original image as the reconstruction of the output image without any operation.The experimental results show that this method can eliminate the noise,and keep agood image details.Finally,the effect is satisfied.