Fingerprints in the form of inked and live-scan fingerprints, have been deployed for human identification for many decades now. With advances in technology and science, fingerprint-based identification has been expanded to the unconstrained domain of law enforce-ment in the form of latent fingerprint based person identification and personal identification on devices based on fingerphoto authentication. To handle capture noise and environmental variations, we observe the need for an efficient representation of fea-tures. This research focuses on developing efficient deep representation learning algorithms for both latent fingerprints and fingerphotos. There are three major contributions of this research: conducting a thorough literature study on the use of deep learning in match-ing controlled and uncontrolled fingerprints, a novel deep group sparse autoencoder (GSAE)-, based latent fingerprint matching algorithm, and a deep ScatNet–based fin-gerphoto matching algorithm. Using these features, we tabulate and compare the results of the proposed algorithms with the existing approaches on the publicly available databases demonstrating improvement in the performance.