Not able to unlock mobile phones while your hands are sweaty can be a real annoyance. Further fingerprint sensors that are typically employed donot do a good job when the finger prints are dirty or smudged. The project aims to overcome this limitation. I've trained a SIAMESE Network on the socofing fingerprint dataset from kaggle and the model has showed some decent results on unseen fingerprints despite the images being artificially tampered to mimic the real world scenarios where the images can't be obtained all polished and neat.
Further the biggest advantage with siamese networks is that once they have been trained they can be used to compare on any new set of images with surprising accuracy, hence the name One shot learning.
Dataset Used: https://www.kaggle.com/ruizgara/socofing
The model took around 10 hours to train on GPU's. The trained model can be directly used to verify for matches among pairs of images.
For further instructions on how the model was trained and the data extracted, check out the two jupyter notebooks from the repo. The model is stored in the saved model format of keras.
Finger_print_verification_part_I.ipynb describes the preprocessing steps that went into the kaggle dataset. Fingerprint_verification_part_II.ipynb contains detals regarding the model training and inference.
Link to trained model and preprocessed images: https://drive.google.com/drive/folders/11-eD3OMVvAx1nvib3cS6O6CUayojdmm1?usp=sharing