This work introduces a dual-branch ensemble framework fusing Semantic Deep Learning with Mathematical Forensic Feature Extraction, highlighting the practicality and scalability of mathematical forensics for real-world deployment.
Abstract
The rapid advancement of text-to-image (T2I) models has necessitated robust Synthetic Image Source Attribution (SIA) methodologies. A critical challenge in SIA is the distribution shift between pristine training images and real-world deployed images, which undergo unknown post-processing operations such as JPEG compression and blurring. In this work, proposed for the DLMMDD Challenge at ICANN 2026, we introduce a dual-branch ensemble framework fusing Semantic Deep Learning with Mathematical Forensic Feature Extraction. The semantic branch employs EfficientNet-B0 regularized with Exponential Moving Averaging (EMA) and Label Smoothing. The forensic branch extracts 126 mathematical features -- including SVD spectral profiles and Local Binary Patterns -- from high-pass noise residuals, compressed via Truncated SVD and classified with XGBoost. Evaluated on a dataset of 10 generators where 55% of the test set is degraded, our approach achieves a private leaderboard accuracy of 95.60%. Furthermore, the entire pipeline is highly computationally efficient, requiring no GPU acceleration and executing end-to-end on a standard CPU in under 6.5 hours, highlighting the practicality and scalability of mathematical forensics for real-world deployment.
This work proposes RNSIDNet, a novel forensic framework that achieves robust detection through enhanced RGB-Noise representation learning, and designs a Hard Sample-aware Contrastive Learning (HSCL) strategy.
Zhen Li, Gang Cao, Tianyi Zhang et al.· 0 citations
A specialized dual-stream framework that strategically integrates features from both spatial and frequency domains to disentangle the fingerprints of heterogeneous generators is proposed, addressing two critical tasks: binary real/fake detection and closed-set source attribution across 10 distinct generative architectures.
L. Pham, Cu Vinh Loc, Truong Nhat Tran et al.· Pattern Analysis and Applica...· 0 citations
The proliferation of hyper-realistic AI-generated images poses significant threats to digital information integrity and forensic accountability. Existing detection methodologies, however, face three critical bottlenecks: vulnerability to real-world distortions such as social media compression, inadequate specialization for socially harmful “ex-regulatory” content, and an inability to perform model attribution essential for effective governance. To address these challenges, we propose the Multi-level Feature Fusion Detection and Attribution Framework (MF2DA), a unified end-to-end pipeline designed for both high-precision detection and reliable model attribution. The core architecture synergizes an Edge Pyramid Fusion ResNet (EPF-ResNet), which captures subtle pixel-level edge artifacts, with a frozen CLIP-ViT to ensure robust semantic generalization. Furthermore, the framework is augmented by the MLLM-Guided Quality Refinement Module (MQRM), which adaptively leverages semantic-agnostic quality features to decouple generative traces from aggressive compression noise. Finally, the Dual-stream Differential Patch Attribution Network (D2PAN) extracts resilient model fingerprints by disentangling micro-textural patterns from semantic interference, thereby achieving precise generator identification. Extensive evaluations on multiple benchmarks demonstrate that MF2DA achieves state-of-the-art performance in detecting both “friendly” and “ex-regulatory” images while maintaining exceptional cross-dataset generalization. By integrating robust detection with precise attribution, this work establishes a practical and accountable forensic solution for the rapidly evolving generative AI landscape.
Wenpeng Mu, Qiang Xu, Yi-Ning Zhang et al.· IEEE Transactions on Informa...· 0 citations
A data-generation pipeline that captions real photographs with a vision–language model and regenerates them with modern text-to-image systems, producing semantically aligned real/synthetic pairs that isolate generative artifacts from image content is described.
In general, the cross-generator generalization and robustness against attacks are two main challenges in AI-generated image detection. To address this, we put forward a synthetic image detector (DTBF), a two-branch architecture followed by a jointly-optimized concatenation (JOC), aiming at capturing low-level artifacts and high-level semantics and dynamically fusing them to enhance the generalization and robustness. In the artifact-extracted branch (AE-branch), the local multi-direction binary-encoding patterns (LMBP) are customized to extract and encode the relationships among pixels within each sliding window, resulting in the LMBP distribution serving as a universal fingerprint to distinguish real/fake images. The dual-alignment guided semantic branch (DAS-branch), working as a supplement to AE-branch, designs global context-unrelated prompts and semantic-enhanced prompts to capture global semantic inconsistency and local patch anomalies, strengthening the generalization and robustness of DTBF. Finally, JOC dynamically fuses the two branches through concatenation to amplify the effectiveness of each branch, achieving better generalization and robustness. With the assistance of two complementary branches and JOC, our proposed DTBF significantly outperforms 12 state-of-the-art detectors on two publicly available datasets in terms of detection accuracy and robustness.
Shaowei Weng, Rui Li, Lifang Yu et al.· IEEE Signal Processing Lette...· 0 citations
The proposed Attention-Based Deep Learning Pipeline of AI-Created Image Recognition incorporates three integrated branches, including low-level statistical feature extraction, high-level semantic representation learning, and attention-based feature refinement mechanism, which support the robustness and generalization ability of the proposed model in detecting AI-generated images in a variety of generators and conditions.
Nadia Ali· Al-Noor Journal of Engineeri...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
