Abstract: Due to the high scattering of near-infrared light in biological tissue, the quality of finger-vein images is often undesirable and the vein network is incomplete.To solve the problem of vascular network coloboma in finger-vein IR images and achieve higher accuracy in finger-vein recognition, a method based on fractal is proposed to restore the finger-vein network. First, the finger-vein images are enhanced by using multi-scale Gabor filters to extract the skeletons of veins. The following steps are based on the skeletons extracted. Second, since part of the vessel segments are lost caused by the binarization in the processing above, the 8-orientation images obtained by Gabor filtering are used to prerestore the vascular structure. Based on the prerestored results, the ratio of branch length to the length of the parent vessel, as a kind of fractal feature, is extracted by using K-means algorithm.Then, the length of the lost blood vessel is calculated based on the fractal feature. Finally, with a movement template of vessel points, the finger vein is simulated by calculating the movement probability of neighbor vessel points. The experimental results show that the proposed method is effective to restore the locally incomplete region of finger-vein network. The proposed method also achieves higher accuracy in finger-vein recognition.