Difference co-occurrence matrix using BP neural network for fingerprint liveness detection

With the growing use of fingerprint identification systems in recent years, preventing fingerprint identification systems from being spoofed by artificial fake fingerprints has become a critical problem. In this paper, we put forward a novel method to detect fingerprint liveness based on BP neural network, which is used for the first time in the fingerprint liveness detection. Moreover, different from traditional detection methods, we propose a scheme to construct the input data and corresponding category labels. More effective and efficient texture features of fingerprints, which are used as the input data of the BP neural network, are computed to improve classification performance and obtain a better pre-trained network model. After a variety of preprocessing operations and image compression operations, gradient values in the horizontal and vertical directions are computed by using Laplacian operator, and difference co-occurrence matrices are constructed from the obtained gradient values. Then, the input data of neural network model are built based on two DCMs. The pre-trained neural network models with diverse neuron nodes are learnt. Different experiments based on different parameters for the BP neural network have been conducted. Finally, classification accuracy of testing fingerprints is predicted based on the pre-trained networks. Experimental results on the LivDet 2013 show that the classification performance of our proposed method is effective and meanwhile provides a better detection accuracy compared with the majority of previously published results.