A Meta-Recognition Based Skin Marks Matching Algorithm with Feature Fusion for Forensic Identification

Abstract

Soft biometrics, such as skin marks, play an important role in forensic identification, for they cannot only supplement hard biometrics to improve the overall identification performance, but may also serve as supportive evidence when hard biometrics is not available. Skin marks are small and difficult to be accurately detected due to different lighting conditions, poses as well as individual variation in their skin marks. In this paper, we propose a meta-recognition based skin marks matching algorithm to address these challenges for forensic identification. The algorithm combines both the geometric information in spatial distribution of skin marks and the appearance information of individual skin mark to establish the correspondence between two images. A multi-level skin marks matching scheme is adopted and fusion of scores is carried out at different levels using a meta-recognition method. The experimental results show that the new algorithm provides over 22% of improvement in terms of rank-1 accuracy over the previously proposed method.