A.I. See You: Leveraging Deep Learning to Triage Pictures and Video

No ratings

Presented at TechnoSecurity&DigitalForensics 2019 by

Matt Mcfadden

There can be hundreds of thousands of images on digital devices – where does an examiner start? Techniques based on file size, date created, and hash sets are a good start, but are insufficient for the sheer volume of data examiners must sort through. What if investigators could view files with specific types of images first? Learning engines can be trained to identify pictures or videos that likely depict categories like weapons, pornography, drugs, and graphic violence. When combining image categorization with previous prioritization techniques, investigators can more quickly triage a device – allowing them to look at the best candidates that may be relevant to their case. Learn how these models are trained, how to appropriately use them, and why it is essential to understand their power and limitations. From assessing which devices need to be seized onsite, to reviewing employee’s adherence to an acceptable use policy, these techniques save time by highlighting for examiners pictures and videos with questionable content.