According to that the acquired finger-vein images contain not only vein features, but also noise and irregular shadows, increasing the difficulty of feature extraction, this paper proposes a novel method for finger-vein image segmentation based on sparse self-encoding. Firstly, the original grayscale image is segmented by the traditional segmentation algorithm to obtain a binary image (the background pixel value is 0 and the vein pixel value is 1). Then, we take each pixel as center point to generate patch from original grayscale image and the value (0 or 1) of the corresponding point in the binary image is used as its label.A training set is built based on generated patches and labels. Finally, the patch images and labels are input into the auto-encoder and neural network for training, and the trained models are used to segment the test images. The experimental results show that compared with the traditional algorithm, the proposed finger vein segmentation algorithm can effectively segment the vein and improve the authentication accuracy of the finger vein authentication system.