A full set of algorithms on image processing and recognition for finger vein authentication was proposed. A new image normalization algorithm was designed. During normalization, a Sobel operator incorporating the first meet tracing was used to detect finger edges, and then image correction with rotation was done after the finger edges were fitted under the least mean-square error principle. Based on maximum curvature models, finger vein was extracted by using four direction gradient operators. To decrease False Rejection Rate (FRR) and increase recognition rapid, after expanded, the image was segmented with three levels and compressed to become a more compact vein pattern including finger vein, background and vague parts. At last, a new matching algorithm with point to point between mask and test patterns was put forward. Our experiments verify that our methods outperform traditional ones on speed and effectiveness. Its recognition rate is up to 100%, with a finger set horizontally.