Wavelet-based features to improve cross-forgery generalization in deepfake detection

Detecting Image Forgeries Using Hybrid Residual-Dense Swin Transformer Framework with Attention-Based U-Net

Establishing the authenticity of digital images is widely acknowledged as a crucial aspect of multimedia security since contemporary alteration techniques like copy-move, splicing, and retouching create remarkably convincing forgeries that are difficult to spot. This research proposes a Deep Learning (DL) based image forgery detection methodology that combines attention-guided segmentation and hybrid feature extraction to deliver robust tamper detection and classification. The preprocessing procedures including resizing and normalizing images are used to ensure consistent model performance. The hybrid encoder, records local texture-level inconsistencies and long-range contextual dependencies, composed of Swin Transformer modules and Residual-Dense blocks. Multi-scale feature maps that are highly discriminative for forgery cues are the outcome of this. These features are decoded using a UNet++ segmentation network that has been altered using Channel-Dual Attention (CA-DA). This improves the spatial focus and boundary precision of the projected forgery mask. The created mask is analyzed in a detection stage, where threshold-based decision-making mechanism determines if the input is genuine or forged, non-empty masks that exceed the threshold indicate tampered images, whereas actual images are represented by empty masks. The proposed system is implemented in Python software, and the results demonstrate that the proposed experimental validation on benchmark datasets indicates improved detection accuracy, robustness to noise compression, and localization clarity when compared to conventional approaches.

Image Forgery Detection Based on Fusion of Lightweight Deep Learning Models

The increasing accessibility of digital image editing tools has made image manipulation easier than ever before, creating significant challenges in verifying the authenticity of digital visual content. Copy-move forgery is one of the most widely used image manipulation techniques because duplicated regions originate from the same image, making manual detection difficult. Existing forgery detection approaches often rely on handcrafted feature extraction or a single deep learning architecture, which may exhibit reduced performance when images undergo transformations such as rotation, scaling, compression, illumination variation, and noise addition. This paper presents a fusion-based lightweight deep learning framework for copy-move image forgery detection and localization. The proposed system integrates three efficient convolutional neural network architectures, namely SqueezeNet, MobileNetV2, and ShuffleNet, to extract complementary deep image features. Instead of depending on a single classifier, the extracted features are fused and classified using a Support Vector Machine (SVM), resulting in improved prediction reliability and reduced false detection. To provide visual evidence of manipulation, the framework incorporates a Discrete Cosine Transform (DCT)-based localization module that identifies duplicated image regions and highlights forged areas. The proposed model was implemented using Python with TensorFlow, OpenCV, Scikit-learn, NumPy, and Tkinter. Experimental evaluation was conducted on the MICC-F220 benchmark dataset. The fusion strategy achieved an overall detection accuracy of 90.91%, outperforming individual lightweight convolutional neural network models while maintaining low computational complexity. The proposed framework offers an efficient and practical solution for digital image authentication and is applicable to digital forensics, journalism, law enforcement, cyber security, and multimedia content verification.

Wtmsa-net: Wavelet transform-based multi-stage attention network for deepfake detection

Generalizable Deepfake Detection via Frequency-Domain Enhancement and Feature Disentanglement

Existing deepfake detectors often perform well on in-domain data but generalize poorly to unseen datasets or manipulation methods. This limitation is largely attributed to their reliance on dataset-specific semantic cues rather than transferable forgery patterns. To address this limitation, we propose a generalizable deepfake detection framework that combines frequency-domain enhancement with feature disentanglement. A Phase-Amplitude Frequency Enhancement (PAFE) module enhances subtle spectral artifacts introduced during deepfake generation. We then feed the enhanced representations into an asymmetric dual-branch architecture that separates content-related information from forgery-related features. The content branch models facial semantics, while the forgery branch extracts discriminative forgery features with reduced content interference. A spatial self-attention module further refines the forgery features. We optimize the framework using image-level reconstruction loss, feature-level contrastive loss, and classification loss. Together, these objectives encourage effective feature disentanglement and improve the discriminability of the learned forgery features. Extensive experiments on several widely used deepfake benchmarks show that the proposed framework achieves competitive detection performance and improved cross-domain generalization compared with existing methods.

Deepfake tampering detection based on multimodal large model

As deep learning generation technologies continue to evolve, Deepfake technology has become a major threat to information security, posing significant challenges to tampering detection. Existing detection methods for deep forgeries generally suffer from insufficient generalization capability and interpretability. To address these issues, this paper proposes a Deepfake tampering detection method based on a multimodal large model, termed LLM-FNP. This method integrates frequency-domain noise perception with visual-language understanding. First, the BayarConv module is employed to extract frequency-domain noise features from images, capturing anomalous noise patterns left by the forgery process, and a dual-branch encoder is used to process the original RGB images and frequency-domain feature maps separately. Second, a cross-attention mechanism is applied to achieve cross-modal fusion. Finally, the fused features are processed by a fine-tuned LLaVA large model, which outputs the detection result along with interpretable text. Experimental results on the DFDCP dataset show that the proposed method achieves an accuracy of 78.54% and an AUC of 81.50%. To evaluate the generalization capability of the method, cross-dataset testing on Celeb-DF and WildDeepfake yields accuracies of 73.5% and 71.8%, and AUCs of 76.4% and 75.1%, respectively, all outperforming classical methods such as ResNet+LSTM, further validating its strong generalization ability.

Enhancing Cross-Dataset Generalization in Image Forgery Detection via Mixed-Domain Training and Adaptive Thresholding

The paper suggests a hybrid training system and adaptive thresholding to improve the generalization of cross-datasets in image forgery detection and indicates that mixed-domain training is a practical approach that can reduce dataset bias and increase generalization.