Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LDR-Net: A Novel Framework for AI-generated Image Detection via Localized Discrepancy Representation

JiaXin Chen, Miao Hu, DengYong Zhang, Yun Song, Xin Liao

TL;DR

AI-generated images increasingly resemble real visuals, challenging authenticity verification. LDR-Net addresses this by fusing Local Gradient Autocorrelation (LGA) and Local Variation Pattern (LVP) to capture smoothing artifacts and irregular pixel variations, using a pruned ResNet classifier for detection. It demonstrates state-of-the-art cross-generator performance on unseen GANs and diffusion models and shows robustness to common post-processing. By focusing on localized, model-agnostic cues rather than generation-specific artifacts, LDR-Net advances practical content authentication in dynamic generation landscapes.

Abstract

With the rapid advancement of generative models, the visual quality of generated images has become nearly indistinguishable from the real ones, posing challenges to content authenticity verification. Existing methods for detecting AI-generated images primarily focus on specific forgery clues, which are often tailored to particular generative models like GANs or diffusion models. These approaches struggle to generalize across architectures. Building on the observation that generative images often exhibit local anomalies, such as excessive smoothness, blurred textures, and unnatural pixel variations in small regions, we propose the localized discrepancy representation network (LDR-Net), a novel approach for detecting AI-generated images. LDR-Net captures smoothing artifacts and texture irregularities, which are common but often overlooked. It integrates two complementary modules: local gradient autocorrelation (LGA) which models local smoothing anomalies to detect smoothing anomalies, and local variation pattern (LVP) which captures unnatural regularities by modeling the complexity of image patterns. By merging LGA and LVP features, a comprehensive representation of localized discrepancies can be provided. Extensive experiments demonstrate that our LDR-Net achieves state-of-the-art performance in detecting generated images and exhibits satisfactory generalization across unseen generative models. The code will be released upon acceptance of this paper.

LDR-Net: A Novel Framework for AI-generated Image Detection via Localized Discrepancy Representation

TL;DR

AI-generated images increasingly resemble real visuals, challenging authenticity verification. LDR-Net addresses this by fusing Local Gradient Autocorrelation (LGA) and Local Variation Pattern (LVP) to capture smoothing artifacts and irregular pixel variations, using a pruned ResNet classifier for detection. It demonstrates state-of-the-art cross-generator performance on unseen GANs and diffusion models and shows robustness to common post-processing. By focusing on localized, model-agnostic cues rather than generation-specific artifacts, LDR-Net advances practical content authentication in dynamic generation landscapes.

Abstract

With the rapid advancement of generative models, the visual quality of generated images has become nearly indistinguishable from the real ones, posing challenges to content authenticity verification. Existing methods for detecting AI-generated images primarily focus on specific forgery clues, which are often tailored to particular generative models like GANs or diffusion models. These approaches struggle to generalize across architectures. Building on the observation that generative images often exhibit local anomalies, such as excessive smoothness, blurred textures, and unnatural pixel variations in small regions, we propose the localized discrepancy representation network (LDR-Net), a novel approach for detecting AI-generated images. LDR-Net captures smoothing artifacts and texture irregularities, which are common but often overlooked. It integrates two complementary modules: local gradient autocorrelation (LGA) which models local smoothing anomalies to detect smoothing anomalies, and local variation pattern (LVP) which captures unnatural regularities by modeling the complexity of image patterns. By merging LGA and LVP features, a comprehensive representation of localized discrepancies can be provided. Extensive experiments demonstrate that our LDR-Net achieves state-of-the-art performance in detecting generated images and exhibits satisfactory generalization across unseen generative models. The code will be released upon acceptance of this paper.
Paper Structure (17 sections, 9 equations, 4 figures, 7 tables)

This paper contains 17 sections, 9 equations, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Local Feature-based Detection
  4. Deep Learning-based Detection
  5. Proposed Method
  6. Local Gradient Autocorrelation
  7. Local Variation Pattern
  8. Classifier
  9. Experiments
  10. Experiment Setup
  11. Ablation Study
  12. GAN-Sources Evaluation
  13. Diffusion-Sources Evaluation
  14. Impact of Hyperparameter Selection
  15. Robustness Evaluation
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Overview of the proposed LDR-Net framework for AI-generated image detection.
  • Figure 2: Illustration of gradient differences between generated images and real images. a) is a fake image generated from the real image b). c) is a scatter plot that statistically represents the gradient magnitude changes between corresponding pixels in a) and b). Our LDR-Net effectively reveals the variation of gradient magnitude between generated images and real images.
  • Figure 3: Comparison of local binary encoding patterns between real and generated images, a) is fake image generated from real image b).
  • Figure 4: CAM visualization for real and generated images.