Vein pattern visualization for CSAM investigation using deep learning

Abstract

ABSTRACT Child sexual abuse material (CSAM) has become one of the fastest-growing criminal industries linked to other illegal activities, such as human trafficking, underage prostitution, and the sextortion of minors. Although common biometric traits such as fingerprint, face, and palmprint are widely used in traditional identification systems, they are ineffective in investigating CSAM cases, as perpetrators often conceal their faces, and only partial non-facial skin may be visible. Vein pattern visualization has been introduced as a new tool in forensic investigations to overcome the limitations of current identification methods. Our early research has shown promising results in uncovering vein patterns from normal digital images using computer vision and deep learning techniques. For this research, we use a dataset of forearms and palm images collected from 301 participants in New Zealand. Utilizing a deep learning framework, we develop a collection of mapping models to reveal unique vein patterns from regular digital images. The results demonstrate our method’s efficiency in visualizing vein patterns, indicating nearly 95% similarity in the contrast values between reference NIR and generated images and 79% in the vein length of the generated and target images.