Criminal and victim identification based on deep and large feature sets from hand biometrics

In forensics, criminal and victim identification based on digital evidence images is highly challenging because the face or other obvious characteristics such as tattoos are occluded, covered, or not visible at all. Existing recognition methods, which make use of biometric characteristics, such as vein, skin mark, height, skin color, weight, race, etc., are considered for solving this problem. The soft biometric traits, including skin color, gender, height, weight and race, provide useful information but not distinctive enough. Veins and skin marks are limited to high-resolution images and some body sites may neither have enough skin marks nor clear veins. Regardless of the availability of these characteristics, other ones, e.g., hands can be used to support the evidence or provide some useful clues to the investigator. Terrorists and rioters tend to expose their hands, including palms and wrists in a gesture of triumph, greeting or salute, while child sexual offenders usually show them when touching their victims. Wrists, in particular, were neglected by the biometric community for forensic applications. To study this problem, a wrist identification algorithm, which includes skin segmentation, key point localization, image to template alignment, large feature set extraction, classification and post-recognition score analysis has been proposed. The proposed algorithm has been evaluated on a new NTU-Wrist-Image-Database-v1, which consists of 3945 images from 731 different wrists, including 205 pairs of wrist images collected from the Internet, taken under uneven illuminations with different poses and resolutions. In the experiments, the proposed algorithm has been compared to palmprint recognition methods. The extracted large feature sets have been studied and compared with selected deep features and feature selection and reduction schemes. The experimental results indicated that wrist is a useful clue for criminal and victim identification. Online palmprint recognition and latent palmprint identification are two branches of palmprint studies. The former uses middle-resolution images collected by a digital camera in a well-controlled or contact-based environment with user cooperation for commercial applications and the latter uses high-resolution latent palmprints collected in crime scenes for forensic investigation. Nevertheless, the merit of contactless palmprint recognition in an uncontrolled and uncooperative environment for forensic investigation is not fully exposed yet. To study this problem, a new palmprint database is established and an end-to-end deep learning algorithm has been proposed. The new database named NTU Palmprints from the Internet (NTU-PI-v1) contains 7881 images from 2035 palms collected from the Internet. The proposed algorithm consists of an alignment network, feature extraction network, in-network data augmentation scheme, and is end-to-end trainable. The proposed algorithm has been compared with the state-of-the-art online palmprint recognition methods and evaluated on three publicly available contactless palmprint databases, IITD, CASIA, and PolyU and two new databases, NTU-PI-v1 and NTU contactless palmprint database. The experimental results showed that the proposed algorithm outperforms the existing palmprint recognition methods. In child sexual abuse images, the chest of the offender can be sometimes visible or even captured simultaneously with the victim's hand. Thus, the nipple-areola complex (NAC) is proposed for offender identification. Popular deep learning and hand-crafted methods are evaluated on a newly established NTU-Nipple-v1 database, which contains 2732 images from 428 different male's NAC. Experimental results indicate that the proposed NAC can be useful for offender identification.Doctor of Philosoph