Developing automated methods to detect and match face and voice biometrics in child sexual abuse videos

Abstract

The proliferation of child sexual abuse material (CSAM) is outpacing law enforcement’s ability to address the problem. In response, investigators are increasingly integrating automated software tools into their investigations. These tools can detect or locate files containing CSAM, and extract information contained within these files to identify both victims and offenders. Software tools using biometric systems have shown promise in this area but are limited in their utility due to a reliance on a single biometric cue (namely, the face). This research seeks to improve current investigative practices by developing a software prototype that uses both faces and voices to match victims and offenders across CSAM videos. This paper describes the development of this prototype and the results of a performance test conducted on a database of CSAM. Future directions for this research are also discussed.