Abstract: Using three-dimensional distance scanner can obtain depth images as well as the corresponding intensity images. Most of the depth images are rough, rich of noise and even incomplete. Therefore, it is easy to generate a 3D bas-relief which has a lot of noise and less of details using depth images-based compression method. In this paper a hybrid algorithm for bas-relief generation of human faces from 3D depth images and 2D intensity images is presented. Firstly, the height field can be compressed to generate the base mesh of bas-relief. Secondly, the intensity images can be used to extract the information of gray level, gradient and saliency, by using saliency information, the gray and gradient information can be added into the height field of base mesh of relief to preserve the facial details. Lastly, smooth the mesh of relief with the Laplace operator on the basis of saliency to make the face smooth. The experimental results show that compared to the depth images-based compression method, the bas-relief generated by our method can keep the facial detailed features much better and compared to the image-based method, our method will avoid the complex interactive operation and keep the whole contour and and shape of human face better.