Fingerprint Generation for DNN Training: A Case Study in Fingerprint Classification
Emre İrtem, Nesli Erdoğmuş · Balkan Journal of Electrical and Computer Engineering · 2025
Large annotated datasets are crucial for training state-of-the-art deep learning systems. However, the availability of publicly accessible fingerprint data significantly lags behind that of image datasets or text corpora, which are extensively utilized for tasks such as image understanding and natural language processing. The challenges associated with the collec-tion and distribution of fingerprint data make synthetic data generation a viable alternative. Nonetheless, existing research primarily focuses on the large-scale evaluation of fingerprint search systems rather than examining the usability of generated fingerprint images for training purposes. This study employs a model-based method to generate synthetic fingerprints and evaluates their effectiveness in training deep neural networks for fingerprint classification. The findings indicate that augmenting the training set with synthetic fingerprint impression images enhances performance comparably to augmenting it with real fingerprint images.