Dr. Benedikt Lorch

How did you first become interested in your field?

As a teenager, I was fascinated by artwork that combined photo editing and 3D modeling to create surreal scenes. Determined to make my own, I started learning to edit photos but soon realized that I needed material to work with. So, I borrowed my parents’ DSLR camera and started experimenting with aperture, shutter speed,  ISO and other settings that could be changed. This sparked my curiosity about how cameras work and how digital images are formed.

Years later, at university, I discovered multimedia forensics, a research field that uses subtle camera acquisition traces to recover information about the origin and authenticity of digital images. Intrigued by these fascinating applications, I leaped at the opportunity to learn from pioneering researchers. I am grateful to many people who nurtured my interest with their insights and encouragement, especially Christian Riess, Felix Freiling, Rainer Böhme, and Hany Farid.

What are you currently working on, and why is it important?

I enjoy building signal processing and learning-based tools to analyze multimedia files, but I also get excited about tracking down the origin of a low-level artifact in a 30-year-old image compression library.

If you had unlimited resources, what problem would you tackle?

If I had unlimited resources and connections to hardware and software manufacturers, I would invest in enhancing active provenance methods for multimedia files.

Methods for verifying the origin and authenticity of multimedia fall into two categories: passive approaches, which look for inconsistencies or fingerprints of specific processing operations in the content itself, and active approaches, some of which attach a cryptographic signature at the time of creation and keep a ledger of all subsequent editing steps. The Coalition of Content Provenance and Authenticity (C2PA) exemplifies an active framework that provides a transparent record of how the file was created and modified.

While it is fascinating how much information can be recovered using passive methods, such analysis is time-consuming and sometimes impossible to give a definite assessment. Active approaches, on the other hand, can fill some of the gaps. I hope that more camera manufacturers, software developers and distribution channels adopt these active provenance solutions. By integrating active provenance solutions, we could make verifying the authenticity of digital content both simpler and more reliable.

What’s the biggest lesson you’ve learned as a researcher?

Details matter, especially in forensic science.

A single inconsistency in the string encoding of an image’s metadata can reveal tampering. A barely noticeable compression pattern can expose that an image was altered from its original recording. A periodic pattern in an image’s noise statistics, a typical artifact in synthetic images, can also appear in natural images because of rounding artifacts in a specific compression library.

The signals forensic methods rely on are often weak and imperceptible. Analyzing these subtle clues requires care and rigor. The more thoroughly you understand the evidence, the more precise conclusions you can draw.

What’s one piece of advice you’d give to aspiring researchers?

Before running an experiment, take the time to think about the expected outcome. If the outcome aligns with your expectations, that is great corroboration. If the outcome diverges from your expectations (and you can exclude the possibility of a bug), you discover something interesting!