In order to overcome the shortcoming of the soft-threshold denosing method in curvelet domain, which may cause edges blurred, a new denoising method based on the edge features of the given image is proposed. In this method, we make full use of the anisotropie advantages of curvelet, extract the detail and edge information from the low-frequency domain and restore it, which can prevent the destruction of the soft-threshold. The results of simulation experiment show that this method can remove the noise and maintain the pictures' edges well. Furthermore, it can improve the value of PSNR, and get better visual effect.