Uploaded July 2022 | Updated September 2026, 7 hours ago
Composite images are the outcome of combining pieces extracted from two or more other images, sometimes with the intent to deceive the observer and convey false narratives. Consider an image suspected of being a composite, and a large corpus of images that might have donated pieces to the composite (such as photos from social media). Inside media forensics, provenance analysis is the problem of (1) finding, within the available corpus, the images that either directly or transitively share content with the composite (namely, the task of provenance retrieval), as well as of (2) establishing the directed acyclic graph whose nodes individually represent the composite and related images, and whose edges express the derivation and content-donation story (e.g., cropping, blurring, splicing) between pairs of images, linking seminal to derived elements (namely, the task of provenance graph construction). In this conversation, we will discuss our most recent advances in provenance analysis, concluding with our latest endeavors towards extending it to unveil disinformation campaigns.
Speakers
Walter Scheirer, Dennis O. Doughty Collegiate Associate Professor, University of Notre Dame
Daniel Moreira, Incoming Assistant Professor, Loyola University
Conversation Leader
Jason Schlessman, Principal Software Engineer, Red Hat
Affiliated Graduate Student
Bill Theisen, PhD student in the Computer Vision Research Lab at the University of Notre Dame
Composite images are the outcome of combining pieces extracted from two or more other images, sometimes with the intent to deceive the observer and convey false narratives. Consider an image suspected of being a composite, and a large corpus of images that might have donated pieces to the composite (such as photos from social media). Inside media forensics, provenance analysis is the problem of (1) finding, within the available corpus, the images that either directly or transitively share content with the composite (namely, the task of provenance retrieval), as well as of (2) establishing the directed acyclic graph whose nodes individually represent the composite and related images, and whose edges express the derivation and content-donation story (e.g., cropping, blurring, splicing) between pairs of images, linking seminal to derived elements (namely, the task of provenance graph construction). In this conversation, we will discuss our most recent advances in provenance analysis, concluding with our latest endeavors towards extending it to unveil disinformation campaigns.
Speakers
Walter Scheirer, Dennis O. Doughty Collegiate Associate Professor, University of Notre Dame
Daniel Moreira, Incoming Assistant Professor, Loyola University
Conversation Leader
Jason Schlessman, Principal Software Engineer, Red Hat
Affiliated Graduate Student
Bill Theisen, PhD student in the Computer Vision Research Lab at the University of Notre Dame
