Segmentation Techniques through Machine Based Learning for Latent Fingerprint Indexing and Identification

Abstract

Latent fingerprints have become most important evidence in law enforcement department and forensic agencies worldwide. It is also very important evidence in forensic applications to identify criminals as it is mostly encountered in crime scenes. Segmentation is one of the solutions to extract quality features. Fingerprint indexing reduces the search space without compromising accuracy. In this paper, minutiae based rotational and translational features and a global matching approach in combination with local matching is used in order to boost the indexing efficiency. Also, a machine learning (ML) based segmentation model is designed as a binary classification model to classify local blocks into foreground and background. Average indexed time as well as accuracy for full as well as partial fingerprints is tabulated by varying the template sminutiae.