The Senior Director of Computer Vision, Anthony Hoogs, Ph.D., recently attended a U.S. Government Video Analytics Conference as an invited speaker, on August 23, 2017. This event, held in the Washington, D.C. area, included three days of talks focused on cutting-edge video analytics, challenges, new techniques, and applications. His talk on “Digital, Physical and Semantic Integrity Assessment on Images and Video” described research from Columbia University, Dartmouth, University of Albany, UC Berkeley, and Kitware, who are all members of Kitware’s team on the DARPA I2O Media Forensics (MediFor) program.
The main focus of this briefing was image and video forensics and the issues that are continuously arising related to fabricated or altered images and video in social media, news, and other outlets. How does one know if an image or video has been altered, manipulated, or fabricated? The Medifor team has developed a suite of visual forensics algorithms to answer these questions, focusing on rebroadcast detection; image cropping detection; video frame dropping; sub-image matching and visual genealogy; reflection detection and statistical manipulation priors, and various other methods. Based on this research and development, they plan to expand on this work including metadata verification and semantic reasoning; wider range video manipulation detection, visual genealogy graph building; object splice detection; and reflection and shadow verification. There is great potential for these types of tools, techniques, and improved capabilities for domains whose focus is to seek, analyze, and execute on valid information. To receive a copy of the pdf presentation, please reach out to firstname.lastname@example.org.