This is a project build by Fang Baole. The CNN model is trained and tested on LivDet 2009, 2011, 2013 and 2015 datasets. It has exceeded the accuracy of SOTA in LivDet 2009 and 2011. Regarding LivDet 2013 and 2015, the model has reached SOTA.
The input of this CNN model is 112 * 112 image which is cropped from the original fingerprint image. Every fingerprint is cropped into 5 112 * 112 pieces. When the model is tested, the detection is based on all five pieces of the orignal fringerprint.
If you have any problems or want to get my datasets and trained model, feel free to contact me by fbl718@sjtu.edu.cn.