Visualising vein pattern using deep learning for forensic investigation

Abstract

Child sexual abuse is a severe global problem that has gained public attention in recent years. Due to the popularity of digital cameras, many perpetrators take images of their sexual activities. Even though most common biometric traits such as the face, DNA, and fingerprint are employed and well established enough to be reliably considered for personal identification purposes, these traits do not apply to child exploitation images, where criminals usually cover their faces, and non-facial skin is partially observable.

The problem of child exploitation is increasing due to the proliferation of such material electronically and the lack of effective identification technology. Thus, alternative methods and biometric traits are needed for identification purposes.

Recently, a new research area, based on non-facial skin features such as vein patterns, has been developed. However, these methods suffer from several weaknesses across different cases, ranging from skin diversity and hairy bodies to subjects with a high volume of dermis fat in which the vein pattern cannot be uncovered.

To use vein patterns in forensic investigation, the authors proposed a new method to extract features using the sparse auto-encoder algorithm and enhance the robustness of vein visualisation by using the denoising auto-encoder algorithm. A pair of synchronised colour and near-infrared (NIR) images are used to generate the skeletonised vein patterns for verifying the outcome of the proposed method and measuring the length of veins by using mean vein diameter. The proposed algorithm was examined on a database with 650 pairs of colour and NIR images and 50 random internet images collected from different body parts such as forearms, thighs, chests and ankles. The experimental results are encouraging and indicate that the potential of using vein patterns in forensic analysis might be practical for forensic investigations.