MLEP: Multi-granularity Local Entropy Patterns for Universal AI-generated Image Detection

View on arXiv ← Back to list

Authors: Lin Yuan, Xiaowan Li, Yan Zhang, Jiawei Zhang, Hongbo Li, Xinbo Gao

Published: 2025-04-18 14:50:23+00:00

AI Summary

This paper proposes Multi-granularity Local Entropy Patterns (MLEP) for AI-generated image detection. MLEP uses entropy feature maps computed from shuffled image patches at multiple scales to capture pixel relationships while mitigating content bias. Extensive experiments demonstrate improved accuracy and generalization compared to state-of-the-art methods.

Abstract

Advancements in image generation technologies have raised significant concerns about their potential misuse, such as producing misinformation and deepfakes. Therefore, there is an urgent need for effective methods to detect AI-generated images (AIGI). Despite progress in AIGI detection, achieving reliable performance across diverse generation models and scenes remains challenging due to the lack of source-invariant features and limited generalization capabilities in existing methods. In this work, we explore the potential of using image entropy as a cue for AIGI detection and propose Multi-granularity Local Entropy Patterns (MLEP), a set of entropy feature maps computed across shuffled small patches over multiple image scaled. MLEP comprehensively captures pixel relationships across dimensions and scales while significantly disrupting image semantics, reducing potential content bias. Leveraging MLEP, a robust CNN-based classifier for AIGI detection can be trained. Extensive experiments conducted in an open-world scenario, evaluating images synthesized by 32 distinct generative models, demonstrate significant improvements over state-of-the-art methods in both accuracy and generalization.

Key findings

MLEP consistently outperforms state-of-the-art methods across various datasets, achieving significant improvements in accuracy and generalization. Ablation studies confirm the importance of patch shuffling and multi-scale analysis in MLEP's effectiveness. Qualitative analysis reveals that MLEP amplifies real-fake discrepancies while minimizing semantic interference.

Approach

The approach uses multi-granularity local entropy patterns (MLEP) computed from shuffled small image patches at multiple scales. These MLEPs are then fed into a CNN classifier to distinguish between real and AI-generated images. The shuffling reduces semantic bias, and multi-scale analysis captures resampling artifacts.

Datasets

Foren-Synths, GANGen-Detection, DiffusionForensics, UniversalFakeDetect, images from Midjourney and DALL-E 2.

Model(s)

ResNet-50 (primarily), ResNet-18, ResNet-34, ResNet-101

Author countries

China

← Previous