Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Analysis of Human Perception in Distinguishing Real and AI-Generated Faces: An Eye-Tracking Based Study

Jin Huang, Subhadra Gopalakrishnan, Trisha Mittal, Jake Zuena, Jaclyn Pytlarz

TL;DR

This study designed a perceptual experiment using eye-tracking technology to analyze how individuals differentiate real faces from those generated by AI, and reveals that participants can distinguish real from fake faces with an average accuracy of 76.80%.

Abstract

Recent advancements in Artificial Intelligence have led to remarkable improvements in generating realistic human faces. While these advancements demonstrate significant progress in generative models, they also raise concerns about the potential misuse of these generated images. In this study, we investigate how humans perceive and distinguish between real and fake images. We designed a perceptual experiment using eye-tracking technology to analyze how individuals differentiate real faces from those generated by AI. Our analysis of StyleGAN-3 generated images reveals that participants can distinguish real from fake faces with an average accuracy of 76.80%. Additionally, we found that participants scrutinize images more closely when they suspect an image to be fake. We believe this study offers valuable insights into human perception of AI-generated media.

Analysis of Human Perception in Distinguishing Real and AI-Generated Faces: An Eye-Tracking Based Study

TL;DR

This study designed a perceptual experiment using eye-tracking technology to analyze how individuals differentiate real faces from those generated by AI, and reveals that participants can distinguish real from fake faces with an average accuracy of 76.80%.

Abstract

Recent advancements in Artificial Intelligence have led to remarkable improvements in generating realistic human faces. While these advancements demonstrate significant progress in generative models, they also raise concerns about the potential misuse of these generated images. In this study, we investigate how humans perceive and distinguish between real and fake images. We designed a perceptual experiment using eye-tracking technology to analyze how individuals differentiate real faces from those generated by AI. Our analysis of StyleGAN-3 generated images reveals that participants can distinguish real from fake faces with an average accuracy of 76.80%. Additionally, we found that participants scrutinize images more closely when they suspect an image to be fake. We believe this study offers valuable insights into human perception of AI-generated media.
Paper Structure (19 sections, 14 figures, 1 table)

This paper contains 19 sections, 14 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Generative Algorithms for Faces
  4. Human Behavioral Study on AI Generated Contents
  5. Human Behavioral Study with Eye-tracking
  6. Study Description
  7. Equipment
  8. Study Design and Setup
  9. Data Summary and Analysis
  10. Data Summary
  11. Recognition Accuracy
  12. Eye-tracking Data
  13. Analysis of Different Attributes
  14. Reaction Time
  15. Mid-point Effect
  16. ...and 4 more sections

Figures (14)

  • Figure 1: Analyzing Human Perception in Distinguishing Real and AI-Generated Faces: We designed a perceptual experiment using eye-tracking to analyze human ability on distinguishing real and fake face images.
  • Figure 2: Experiment Setup: (left) Participants were seated 64" away from the TV screen (55 ") that was used to show them stimuli (real/fake face images). The eyetracker was placed 20" from the participant at an angle of 31 degrees. Participants used a keyboard for the experiment. (right) A mock-up of the stimulus presentation to participants.
  • Figure 3: Participant-wise Recognition Accuracy: For each participant, the blue bar presents the recognition accuracy on all images; the orange bar presents the recognition accuracy on real images, and the green bar presents the recognition accuracy on fake images. Data is sorted in ascending order by overall accuracy and participant indices are reset after sorting.
  • Figure 4: Confusion Matrix : Analysis of cumulative participants performance across all real and fake images.
  • Figure 5: Gaze Heatmaps: Real and fake face images and gaze heatmaps. Red/orange regions had longer fixation duration than blue/green regions.
  • ...and 9 more figures