The Statistical Footprint of AI-Generated Images

Introduction

The Blur Between Real and Synthetic

Figure 1: "The Electrician" by Boris Eldagsen, the AI-generated image that won the Creative category at the 2023 Sony World Photography Awards.

In 2023, the Sony World Photography Awards—one of the most prestigious competitions in the field—awarded a prize in the Creative category to an image titled "The Electrician". The image presented a haunting, black-and-white portrait of two women. However, the artist, Boris Eldagsen, refused the award, revealing that the work was not a photograph at all, but a synthetic creation generated by AI. This event marked a significant turning point: synthetic imagery had crossed a threshold of fidelity where even expert judges could no longer distinguish pixels captured by photons from pixels hallucinated by neural networks.<ref>Insert citation for Sony World Photography Awards / Boris Eldagsen controversy here.</ref>

The Scale of Generation

Figure 2: Statistics comparing the timeline of AI image generation vs. traditional photography. <ref name="everypixel">Everypixel Journal, "AI Image Statistics Report," August 2023.</ref>

This blurring of reality is compounded by the unprecedented scale of production. Recent reports indicate that in just 1.5 years, generative AI models have produced as many images as traditional photography produced in its first 150 years.<ref name="everypixel" /> We are rapidly entering an era where a significant portion of digital visual data is synthetic.

The Research Question

For the general public, the inability to distinguish real from synthetic is a question of misinformation. However, for Image Systems Engineering, it is a question of safety.

Autonomous Vehicle (AV) developers are increasingly turning to generative AI to create training data for perception systems to bridge the gap between expensive real-world data and the need for massive edge-case datasets. This poses a critical question: "Can we discern the real from the artificial?"

If these "Digital Twins" appear realistic to human observers but fail to behave like real sensors—lacking the correct noise, spectral, or optical properties—they risk introducing a Domain Gap where AVs are trained on hallucinations rather than physics.

Project Goals

This project conducts a forensic analysis of AI-generated sensor data to determine its viability for simulation. We ignore the semantic content (e.g., whether the car looks like a car) and focus entirely on the physical statistics (how the image was formed).

By comparing a Physical Ground Truth (simulated via ISETCam) against a Generative Reconstruction (generated via Stable Diffusion v1.5), we aim to quantify the statistical "fingerprints" of the AI across four domains:

• Spatial Statistics: Texture and frequency distribution.
• Photometric Statistics: Signal-dependent noise response.
• Spectral Statistics: Inter-channel color correlation.
• Optical Statistics: Point spread function (PSF) and diffraction.