Abstract:Currently, with the continuous advancement of deep learning technology, numerous neural network-based image tampering detection models have emerged, achieving promising detection results. Traditional algorithms based on handcrafted features are gradually being replaced. However, these models often exhibit unstable detection accuracy for different types of tampered images due to their reliance on feature extraction techniques like filters and constrained convolutions. Furthermore, these models generally lack generalization capability and fail to detect previously unseen types of tampering. In recent years, emerging technologies such as attention mechanisms have been widely applied and researched in the field of image processing. Attention mechanisms allow models to focus on critical parts of an image, specifically the tampered regions, significantly improving the accuracy of image tampering detection tasks. To address the aforementioned issues and enhance model detection accuracy, this thesis proposes a multi-feature fusion network based on attention mechanisms. An adaptive feature extraction module is designed to integrate image RGB, noise, and edge information, applying attention mechanisms in both spatial and channel dimensions. This attention-guided enhanced neural network enables the spontaneous learning of tampering features, facilitating the detection of various types of tampering. Additionally, a low-level pixel branch is incorporated to enable the network to notice tampering traces in low-level texture information, further improving detection accuracy. Experimental results demonstrate that, compared to existing detection models, the proposed model not only achieves superior detection accuracy but also exhibits greater robustness.

Two-Stream Neural Networks for Tampered Face Detection

Hierarchical Supervisions with Two-Stream Network for Deepfake Detection.

A Cascade Face Spoofing Detector Based on Face Anti-Spoofing R-CNN and Improved Retinex LBP

Digital image manipulation detection with weak feature stream

Attention-Based Two-Stream Convolutional Networks for Face Spoofing Detection

Interactive Two-Stream Network Across Modalities for Deepfake Detection

Combined spatial and frequency dual stream network for face forgery detection

Two-stream Convolutional Networks for Multi-frame Face Anti-spoofing

SRTNet: a spatial and residual based two-stream neural network for deepfakes detection

Learning Rich Features for Image Manipulation Detection

Image Tampering Detection Method Based on Multi-Feature Fusion

Document images forgery localization using a two‐stream network

Deep Frequent Spatial Temporal Learning for Face Anti-Spoofing

MesoNet: a Compact Facial Video Forgery Detection Network

Spatial-temporal Transformer Network for Protecting Person-of-interest from Deepfaking

Face Forgery Detection Based on the Improved Siamese Network

Two-stream Encoder-Decoder Network for Localizing Image Forgeries

DMFF-Net: Double-stream multilevel feature fusion network for image forgery localization

A Two-Stage Dual-Path Framework for Text Tampering Detection and Recognition

A Dual-channel Augmented Attentive Dense-convolutional Network for power image splicing tamper detection

FGDNet: Fine-Grained Detection Network Towards Face Anti-Spoofing