# Humans Are Easily Fooled by Digital Images

arXiv: https://arxiv.org/abs/1509.05301 · HTML: https://mlsanity.com/paper/1509.05301/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Sep 2015.

## Built on by (most cited first, 8 of 8)

- [ForgerySleuth: Empowering Multimodal Large Language Models for Image Manipulation Detection](https://mlsanity.com/paper/2411.19466/) (2024)
- [Fusion Transformer with Object Mask Guidance for Image Forgery Analysis](https://mlsanity.com/paper/2403.12229/) (2024)
- [Unmasking Illusions: Understanding Human Perception of Audiovisual Deepfakes](https://mlsanity.com/paper/2405.04097/) (2024)
- [Distinguishing Natural and Computer-Generated Images using Multi-Colorspace fused EfficientNet](https://mlsanity.com/paper/2110.09428/) (2021)
- [Exploring Saliency Bias in Manipulation Detection](https://mlsanity.com/paper/2402.07338/) (2024)
- [Towards a New Science of Disinformation](https://mlsanity.com/paper/2204.01489/) (2022)
- [Deepfake Detection using ImageNet models and Temporal Images of 468 Facial Landmarks](https://mlsanity.com/paper/2208.06990/) (2022)
- [An Independent Discriminant Network Towards Identification of Counterfeit Images and Videos](https://mlsanity.com/paper/2506.05377/) (2025)

## Abstract (arXiv)

Digital images are ubiquitous in our modern lives, with uses ranging from social media to news, and even scientific papers. For this reason, it is crucial evaluate how accurate people are when performing the task of identify doctored images. In this paper, we performed an extensive user study evaluating subjects capacity to detect fake images. After observing an image, users have been asked if it had been altered or not. If the user answered the image has been altered, he had to provide evidence in the form of a click on the image. We collected 17,208 individual answers from 383 users, using 177 images selected from public forensic databases. Different from other previously studies, our method propose different ways to avoid lucky guess when evaluating users answers. Our results indicate that people show inaccurate skills at differentiating between altered and non-altered images, with an accuracy of 58%, and only identifying the modified images 46.5% of the time. We also track user features such as age, answering time, confidence, providing deep analysis of how such variables influence on the users' performance.