Preprocessing Genuine and Fake Fingerprint Images and Recognition Based on Neural Network

Abstract

Fingerprint feature information can be used for individual identification. The appearance of forged fingerprints has a negative impact on the authenticity of individual identification. In this paper, a method based on the neural network is proposed to identify the genuine and fake fingerprint images. The method preprocesses the fingerprint images. The resolution of each fingerprint image is set to 500 dpi. Then, the fingerprint images are cut. A moving window with the size of 360 × 256 is defined to move the window on the fingerprint image at certain intervals. The proportion of the effective fingerprint area in the moving window is calculated. When the proportion of the effective fingerprint area reaches or exceeds a threshold, the color subimage in the moving window is saved as a training sample or a test sample. It is necessary to normalize the mean and variance of the fingerprint image before the fingerprint image is inputted into the neural network. The neural network proposed in this paper is based on the residual neural modules. The neural network is composed of a convolutional layer, a max-pooling layer, four residual modules and three fully-connected layers. The cross-entropy loss function is used as the objective function of the neural network. Adam algorithm is employed to optimize the parameters of the neural network. The proposed method is evaluated in different training sample datasets and test sample datasets which include the genuine and fake fingerprint images. The neural network method is compared with the k-nearest neighbor method in identifying the genuine and fake fingerprint images. The experimental results show that the method is superior to the k-nearest neighbor method in the accuracy of identifying the genuine and fake fingerprint images.