Deep Learning Based Latent Palmprint Recognition

Biometric data is utilized by governmental and institutional entities to ascertain the unique identity of an individual. Palm print data, a significant element of biometric information, is frequently employed in applications such as forensic investigations, serving a critical role in the ascertainment of individual identity. The development of automated palm print recognition algorithms presents significant challenges due to the noise introduced during palm print acquisition processes and the inherent complexity of palm feature extraction. In this research, a deep learning-driven methodology for palm print recognition is introduced. The proposed method is capable of processing both crime scene-acquired and sensor-derived palm print data. The palm print matching process operates within Euclidean space, utilizing vectors generated from minutiae segments that undergo a series of specific stages. The vectors corresponding to matched minutiae, when processed through the trained deep learning architecture, exhibit proximity to each other within the Euclidean space, while non-corresponding minutiae vectors display significant distance. The performance of the proposed methodology has been tested both publicly available and private datasets.