AI-Enabled Visual Evidence Intelligence for Detecting Manipulated Digital Media

Abstract

The availability of sophisticated tools for creating and manipulating digital media content has fueled the proliferation of digital media manipulation. Manipulated digital media (MDM), especially in the form of synthetic and spliced images/videos, are becoming a pervasive threat to the credibility of visual evidence. MDM detections have implications on a wide range of applications, spanning news/media, law enforcement, national security, and the judiciary. Despite ongoing research efforts, the need for a comprehensive solution to the detection of such MDM remains a critical challenge for the academic community and stakeholders.

While MDM detection draws substantial attention, there is currently no Visual Evidence Intelligence (VEI) that enables evidence-based conclusions on the verifiability/authenticity of images/videos, the veracity of visual claims made about the items, or the sufficiency of supporting evidence and cues for facilitating such conclusions. The absence of VEI hinders society’s ability to detect sophisticated MDM that evade state-of-the-art detection systems. Addressing this need necessitates a shift from the traditional approach of establishing ad-hoc detectors for specific categories of MDM towards comprehensive end-to-end research that emulates how humans use visual evidence to make conclusions about images/videos.