Abstract: Traditional leaf carving requires complicated handwork, which is not easily inherited to the next generation. In this paper, we present a method to construct 3D leaf carving from single image. Firstly, we divide the image into three parts, the hollow part, the useless part and the useful part. Then, according to user’s selection, the slender petiole is extracted by traversing pixels. After that, we transform the image into distance field, construct the initial 3D model, and refine it using sigmoid function. Gray scale morphology method is employed to extract the leaf vein and add feature details. The 3D model is smoothed by feature preserving smooth algorithm. Finally, we obtain the skeleton points in narrow parts using the method similar to fluid diffusion and do a process of connecting, extending and triangle structure reconstruction to avoid dimple phenomena. Based on model constraints, we get boundary points. To adjust the 3D model, the heights of boundary points, skeleton points in narrow parts and petiole skeleton points are reconnected. The final model is generated by inserting points to repaire breaking parts. Experimental results show that the proposed method can generate good 3D leaf carving model.