Construction and Research of Neural Network Model in Fingerprint Comparison

In fingerprint image processing, the traditional fingerprint recognition algorithm has a low fingerprint recognition rate due to the blur, damage and loss of fingerprints. The existing deep neural network-based fingerprint recognition method model has a large number of parameters. In order to solve the above problems, the FPRNet model based on convolutional neural network is proposed in this paper. The subtraction method of fingerprint feature vector is adopted to extract the quadratic feature of the subtraction results, which improves the accuracy of fingerprint recognition. After preprocessing the original fingerprint, such as filtering enhancement, gray image processing and binarization, the algorithm inputs the preprocessed fingerprint image into the FPRNet model for feature extraction, and subtracts the two groups of feature vectors. The obtained feature vectors are input into the next layer network, and the output feature vectors can be compared to determine whether the two fingerprints are the same. The experiment of FPRNet model shows that the fingerprint recognition rate is 96.23% through this model without image preprocessing, and 98.72% through this model after fingerprint image preprocessing. And the FPRNet model size is only 29.2MB. Then, the FPRNet model is compared with Inception V3 model, NASNet Large model and ResNet50 model. Experimental data show that this model has a higher generalization ability for the unpreprocessed fingerprints, a higher recognition rate than other algorithms, and a higher recognition efficiency than the existing network models.