CoChat
Sign Up
Cited by 14Digital Media Forensic DetectionGenerative Adversarial Networks and Image SynthesisAdvanced Neural Network Applications

Vision Transformers (ViTs) for Feature Extraction and Classification of AI-Generated Visual Designs

Open access
Yun, Qing·2025·IEEE Access
DOI 10.1109/access.2025.3562130
Open this paper in CoChat — free account

Abstract

Deep learning has become a cornerstone of modern Artificial Intelligence (AI), enabling machines to process and interpret complex visual information with unprecedented accuracy. As AI-generated content becomes more realistic, the ability to distinguish between machine-created and human-created images is increasingly important. This challenge extends beyond technical concerns, influencing digital media credibility, intellectual property rights, and the integrity of visual communication. Developing robust classification models to accurately attribute image origins is crucial for ensuring transparency, preventing misinformation, and upholding artistic authenticity in an era of rapidly evolving generative AI technologies. This study addresses the critical need to differentiate between AI-generated and human-generated aesthetic images through the application of advanced deep learning models. We investigate the effectiveness of advanced deep learning architectures including High Resolution Networks (HRNet), and Vision Transformers (ViT) which are generally accurate when used to infer the creative characteristics of human visual art. The proposed model ViT, employing a mechanism of self-attention to process images as sequences of patches for feature extraction, is examined for its potential to capture global contextual relationships within images, which is essential for recognizing the nuanced differences between AI and human artistry. ViT achieves 97% accuracy shows that superior performance validates its ability, using its transformer structure, to analyze and learn about the complex features of images which disclose their origin as compared to HRNet model of 95%. This research highlights the potential of using sophisticated deep learning techniques to address the challenges of content authenticity in digital media. By leveraging the unique strengths of each model, we provide insights into their applicability and effectiveness in distinguishing between different forms of digital creation, marking a significant step forward in the field of digital forensics and content verification.

Integrity signals

Registry

Crossref · 10.1109/access.2025.3562130

Cites

39 of 39 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Claims the paper makes

We couldn’t retrieve this paper’s full text, so its claims and self-check weren’t run. The registry and citation checks still apply.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Transformer-based models achieve notable accuracy improvements, particularly for low-quality forgeries, but their computational intensity remains a significant barrier to real-time deployment.

    “While these approaches achieve notable accuracy improvements, particularly for low-quality forgeries, their computational intensity remains a significant barrier to real-time deployment.”

    From Beyond Accuracy: A Deployment-Oriented Benchmark of Deepfake Detection Models (2025) · cites this paper as [10]

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Yun, Q.
Affiliations
Kyungil University (KR)
Corresponding author
Yun, Qing
ORCID
Yun, Qing
Venue
IEEE Access
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2025
Volume
13
Pages
69459-69477
ISSN
2169-3536 (ISSN-L)
DOI
10.1109/access.2025.3562130
OpenAlex
W4409560444
Cited by
13
References
39
Citations by year
2026: 9 · 2025: 4
FWCI
4.60
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Access (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Keywords
AI-generated image detection, vision transformer, feature extraction, image classification, self-attention, HRNet, generative AI, content authenticity, digital forensics, content verification

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review