Lifang Yu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Lifang Yu has a total of 52 co-authors. Here are the top co-authors by number of shared publications:
- Shaowei Weng
Fujian University of Technology
24 shared publications - Gang Cao
School of Computer and Cyber Sciences, Communication University of China, Beijing, China
10 shared publications - Chunyu Zhang
Xizang Minzu University
3 shared publications - Jian He
School of Electronic, Electrical Engineering and Physics, Fujian University of Technology, Fuzhou, China
3 shared publications - Li Li
School of Computer Science and Technology, Hangzhou Dianzi University, Hangzhou 310018, China
3 shared publications - Liwen Chen
Fujian Provincial Key Laboratory of Big Data Mining and Applications, Fujian University of Technology, Fuzhou, China
3 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (11) , 17455-17469
KDGW-Frame: A Robust Keypoint-Dominated Watermarking Framework With Accurate Localization and Rectification
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (11) , 17455-17469
KDGW-Frame: A Robust Keypoint-Dominated Watermarking Framework With Accurate Localization and Rectification
- 29 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
LGC-END: A Dual-Branch Architecture Achieving the Fusion of the Watermark and the Cover Image from Multiple Perspectives
The two most important properties of image watermarking techniques are to ensure high imperceptibility of watermarked images, and to enhance the extraction accuracy of the embedded watermark when facing varying attacks. Actually, striking a satisfactory balance between the two properties under various attacks remains a great challenge to image watermarking. To address this challenge, a local-global collaborative encoder-noiselayer-decoder (END) framework, named LGC-END, is proposed in this paper to achieve comprehensive integration of the watermark and the cover image from local-global and spatial-frequency perspectives. The encoder/decoder adopt similar local-global collaborative architecture to realize high-quality watermark embedding and high-accuracy watermark extraction under various attacks. In the encoder, the local branch utilizes a set of parallel depth-wise separable convolutions to grasp multi-scale contextual information. A multi-frequency channel attention (MFCA) module is proposed to first transform one channel of each scale feature map into multiple frequency components, and then model the importance of channels based on the weighted frequency energy of each channel, highlighting the channels suitable for watermark embedding while suppressing other channels. Subsequently, a multi-scale interactive spatial attention (MSISA) module is designed to capture the spatial commonality among three scale feature maps and strengthen the spatial positions of each scale feature map, which are suitable for watermark embedding, using the extracted commonality. The three-stage global branch is introduced to model the long-range dependencies of the input features stage by stage, allowing each bit of the watermark to diffuse a wider range of image regions and improving the capability of resisting the geometric attacks. A large number of experimental results have shown that the proposed LGC-END is significantly superior to existing advanced methods in various distortion scenarios, and has superior robustness while maintaining high visual quality of images.
- 29 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
LGC-END: A Dual-Branch Architecture Achieving the Fusion of the Watermark and the Cover Image from Multiple Perspectives
The two most important properties of image watermarking techniques are to ensure high imperceptibility of watermarked images, and to enhance the extraction accuracy of the embedded watermark when facing varying attacks. Actually, striking a satisfactory balance between the two properties under various attacks remains a great challenge to image watermarking. To address this challenge, a local-global collaborative encoder-noiselayer-decoder (END) framework, named LGC-END, is proposed in this paper to achieve comprehensive integration of the watermark and the cover image from local-global and spatial-frequency perspectives. The encoder/decoder adopt similar local-global collaborative architecture to realize high-quality watermark embedding and high-accuracy watermark extraction under various attacks. In the encoder, the local branch utilizes a set of parallel depth-wise separable convolutions to grasp multi-scale contextual information. A multi-frequency channel attention (MFCA) module is proposed to first transform one channel of each scale feature map into multiple frequency components, and then model the importance of channels based on the weighted frequency energy of each channel, highlighting the channels suitable for watermark embedding while suppressing other channels. Subsequently, a multi-scale interactive spatial attention (MSISA) module is designed to capture the spatial commonality among three scale feature maps and strengthen the spatial positions of each scale feature map, which are suitable for watermark embedding, using the extracted commonality. The three-stage global branch is introduced to model the long-range dependencies of the input features stage by stage, allowing each bit of the watermark to diffuse a wider range of image regions and improving the capability of resisting the geometric attacks. A large number of experimental results have shown that the proposed LGC-END is significantly superior to existing advanced methods in various distortion scenarios, and has superior robustness while maintaining high visual quality of images.
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 818-822
Transferable Dual-Domain Feature Importance Attack Against AI-Generated Image Detector
Recent AI-generated image (AIGI) detectors achieve impressive accuracy under clean condition. In view of anti-forensics, it is significant to develop advanced adversarial attacks for evaluating the security of such detectors, which remains unexplored sufficiently. This letter proposes a Dual-domain Feature Importance Attack (DuFIA) scheme to invalidate AIGI detectors to some extent. Forensically important features are captured by the spatially interpolated gradient and frequency-aware perturbation. The adversarial transferability is enhanced by jointly modeling spatial and frequency-domain feature importances, which are fused to guide the optimization-based adversarial example generation. Extensive experiments across various AIGI detectors verify the cross-model transferability, transparency and robustness of DuFIA.
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 818-822
Transferable Dual-Domain Feature Importance Attack Against AI-Generated Image Detector
Recent AI-generated image (AIGI) detectors achieve impressive accuracy under clean condition. In view of anti-forensics, it is significant to develop advanced adversarial attacks for evaluating the security of such detectors, which remains unexplored sufficiently. This letter proposes a Dual-domain Feature Importance Attack (DuFIA) scheme to invalidate AIGI detectors to some extent. Forensically important features are captured by the spatially interpolated gradient and frequency-aware perturbation. The adversarial transferability is enhanced by jointly modeling spatial and frequency-domain feature importances, which are fused to guide the optimization-based adversarial example generation. Extensive experiments across various AIGI detectors verify the cross-model transferability, transparency and robustness of DuFIA.
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (6) , 8717-8731
DCNet: Learning Similarity and Spatial Complementary Features for Generalized AI-Generated Image Detection
New generators for producing fake images are always coming up, which poses a serious challenge for developing a detector with strong generalization ability. This paper notices this challenge and presents a dual-branch cross-stage interactive detector, called DCNet, which is comprised of preprocessing, two parallel three-stage branches, and two cross-branch agent attention fusion (CBAF) modules placed between the two branches. The primary purpose of DCNet is to enhance the generalization ability by utilizing the two branches to grasp and refine periodic patterns exhibited by upsampling artifacts as well as the local details of upampling artifacts, respectively. A multi-dilation similarity extraction module (MSEM) is placed at the beginning of the upper branch to capture the periodic grid-like patterns of different sizes exhibited by upsampling artifacts by innovatively incorporating the cosine similarity at varying dilation rates. Simultaneously, under the guidance of MSEM, a similarity-guided spatial artifact attention module (SSAM) is deployed at the beginning of the bottom branch to extract local contextual features related to upsampling artifacts using standard convolutions. Each CBAF is tailored to collect global contextual information from both branches at low computational cost, guide them to learn from each other to filter out the discriminative features, and provide the feedback to each branch to facilitate their respective feature refinement. The local-global-guided convolution (LGConv) in the 2nd and 3rd stage is designed to generate adaptive convolutional kernel weights and position weights for each spatial position of the input feature map by means of local multi-scale features from the previous stage and the global contextual features from CBAF, thereby enhancing key information related to artifacts in deep layers of DCNet. Extensive experimental results demonstrate that the proposed DCNet significantly outperforms existing detectors in terms of generalization capability across various unseen generation models.
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (6) , 8717-8731
DCNet: Learning Similarity and Spatial Complementary Features for Generalized AI-Generated Image Detection
New generators for producing fake images are always coming up, which poses a serious challenge for developing a detector with strong generalization ability. This paper notices this challenge and presents a dual-branch cross-stage interactive detector, called DCNet, which is comprised of preprocessing, two parallel three-stage branches, and two cross-branch agent attention fusion (CBAF) modules placed between the two branches. The primary purpose of DCNet is to enhance the generalization ability by utilizing the two branches to grasp and refine periodic patterns exhibited by upsampling artifacts as well as the local details of upampling artifacts, respectively. A multi-dilation similarity extraction module (MSEM) is placed at the beginning of the upper branch to capture the periodic grid-like patterns of different sizes exhibited by upsampling artifacts by innovatively incorporating the cosine similarity at varying dilation rates. Simultaneously, under the guidance of MSEM, a similarity-guided spatial artifact attention module (SSAM) is deployed at the beginning of the bottom branch to extract local contextual features related to upsampling artifacts using standard convolutions. Each CBAF is tailored to collect global contextual information from both branches at low computational cost, guide them to learn from each other to filter out the discriminative features, and provide the feedback to each branch to facilitate their respective feature refinement. The local-global-guided convolution (LGConv) in the 2nd and 3rd stage is designed to generate adaptive convolutional kernel weights and position weights for each spatial position of the input feature map by means of local multi-scale features from the previous stage and the global contextual features from CBAF, thereby enhancing key information related to artifacts in deep layers of DCNet. Extensive experimental results demonstrate that the proposed DCNet significantly outperforms existing detectors in terms of generalization capability across various unseen generation models.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28 (15209210) , 415-425
DV-Net: Detecting and Distinguishing Copy-Move Regions From Dual Views
Copy-move forgery detection (CMFD) is a technique tailored to detect the existence of copy-move regions in a query image. In this paper, a dual-view CMFD network named DV-Net is proposed, which integrates the combination of the similarity information and tampered features from shallow features conducive to copy-move region localization by using dual-view self-correlation calculation (DV-SCC) and the shallow similarity attention module (SSAM), and strengthens the ability of distinguishing source/target regions by making deep features pass through three serial multiple serial adaptive receptive field selection modules (ARFSMs). The SCC plays an irreplaceable role in identifying copy-move regions. However, single-view SCC, such as the cosine similarity or the Euclidean distance, can solely capture the similarly information from a single perspective. DV-SCC, a combination of Euclidean distance and cosine similarity, provides more comprehensive similarity information from numerical and directional perspectives. In addition, different from previous CMFD networks that only utilize the similarity information to locate similar regions while neglecting tampered features contained in the shallow features, which are of vital importance to CMFD, we innovatively convert the similarity information into the SSAM and apply SSAM on the shallow features to emphasize the similarity information while preserving tampered features, significantly enhancing the localization accuracy of source/target regions. Multiple serial ARFSMs, each containing two parallel branches controlled by a soft attention, can adaptively select appropriate receptive fields according to the scales of tampered regions, improving the classification accuracy of source/target regions. The experimental results show that DV-Net outperforms several advanced algorithms in source/target region localization and discrimination on three publicly available datasets.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28 (15209210) , 415-425
DV-Net: Detecting and Distinguishing Copy-Move Regions From Dual Views
Copy-move forgery detection (CMFD) is a technique tailored to detect the existence of copy-move regions in a query image. In this paper, a dual-view CMFD network named DV-Net is proposed, which integrates the combination of the similarity information and tampered features from shallow features conducive to copy-move region localization by using dual-view self-correlation calculation (DV-SCC) and the shallow similarity attention module (SSAM), and strengthens the ability of distinguishing source/target regions by making deep features pass through three serial multiple serial adaptive receptive field selection modules (ARFSMs). The SCC plays an irreplaceable role in identifying copy-move regions. However, single-view SCC, such as the cosine similarity or the Euclidean distance, can solely capture the similarly information from a single perspective. DV-SCC, a combination of Euclidean distance and cosine similarity, provides more comprehensive similarity information from numerical and directional perspectives. In addition, different from previous CMFD networks that only utilize the similarity information to locate similar regions while neglecting tampered features contained in the shallow features, which are of vital importance to CMFD, we innovatively convert the similarity information into the SSAM and apply SSAM on the shallow features to emphasize the similarity information while preserving tampered features, significantly enhancing the localization accuracy of source/target regions. Multiple serial ARFSMs, each containing two parallel branches controlled by a soft attention, can adaptively select appropriate receptive fields according to the scales of tampered regions, improving the classification accuracy of source/target regions. The experimental results show that DV-Net outperforms several advanced algorithms in source/target region localization and discrimination on three publicly available datasets.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 16772-16783
Large-Capacity Robust Video Watermarking Against Geometric Attacks via CV-VAE Latent Space Compression
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 16772-16783
Large-Capacity Robust Video Watermarking Against Geometric Attacks via CV-VAE Latent Space Compression
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 2834-2838
DTBF: Combining Local Statistical Artifacts and Concept Alignment for Synthetic Image Detection
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 2834-2838
DTBF: Combining Local Statistical Artifacts and Concept Alignment for Synthetic Image Detection
- 2 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. PP (10709908) , 1-5
WL-WEM Combining Low-Cost Watermark Enhancement Modules for In-Generation Watermarking
In general, modifying the latent diffusion model (LDM) decoder to achieve in-generation watermarking cannot introduce tremendous computational burden, which easily leads to non-convergence. This necessarily increases the difficulty of embedding the watermark into the LDM decoder due to the need to strike a balance among imperceptibility, robustness and computational cost. We realize the difficulty and design two lightweight watermarking modules, namely a low-cost watermark redundancy enhancement module (WREM) and a latent-guided watermark enhancement module (LWEM), aiming at reducing the modifications to the LDM decoder as much as possible while maintaining the generation quality and enhancing the robustness. Specifically, WREM, specially designed for shallow layers, utilizes a small number of repetition operations to strengthen the robustness of the watermark, and adopts a low-cost sub-pixel convolution layer to achieve dimension consistency between the watermark residual and the input latent, greatly reducing the computational cost while enhancing the integration of watermark features and the latent feature. LWEM, tailored for deep layers, innovatively exploits a simple bilinear interpolation to strengthen the robustness of the watermark, and fuses watermark features and the latent feature using a cheap convolution layer so as to generate the watermark residual with relatively low impact on the input latent. Combining WREM and LWEM, we construct a lightweight encoder-noiselayer-decoder in-generation watermarking method dubbed WL-WEM pursuing a satisfactory balance among three metrics including computational cost, generation quality and robustness. Experimental results also demonstrate that the proposed WL-WEM outperforms several related works in balancing three metrics.
- 2 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. PP (10709908) , 1-5
WL-WEM Combining Low-Cost Watermark Enhancement Modules for In-Generation Watermarking
In general, modifying the latent diffusion model (LDM) decoder to achieve in-generation watermarking cannot introduce tremendous computational burden, which easily leads to non-convergence. This necessarily increases the difficulty of embedding the watermark into the LDM decoder due to the need to strike a balance among imperceptibility, robustness and computational cost. We realize the difficulty and design two lightweight watermarking modules, namely a low-cost watermark redundancy enhancement module (WREM) and a latent-guided watermark enhancement module (LWEM), aiming at reducing the modifications to the LDM decoder as much as possible while maintaining the generation quality and enhancing the robustness. Specifically, WREM, specially designed for shallow layers, utilizes a small number of repetition operations to strengthen the robustness of the watermark, and adopts a low-cost sub-pixel convolution layer to achieve dimension consistency between the watermark residual and the input latent, greatly reducing the computational cost while enhancing the integration of watermark features and the latent feature. LWEM, tailored for deep layers, innovatively exploits a simple bilinear interpolation to strengthen the robustness of the watermark, and fuses watermark features and the latent feature using a cheap convolution layer so as to generate the watermark residual with relatively low impact on the input latent. Combining WREM and LWEM, we construct a lightweight encoder-noiselayer-decoder in-generation watermarking method dubbed WL-WEM pursuing a satisfactory balance among three metrics including computational cost, generation quality and robustness. Experimental results also demonstrate that the proposed WL-WEM outperforms several related works in balancing three metrics.
- 5 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 22 (6) , 7215-7228
Trusted Video Inpainting Localization via Deep Attentive Noise Learning
Digital video inpainting technique has been substantially improved with deep learning in recent years. It may be used as malicious manipulation to remove important objects for creating forged videos. As such it is significant to blindly identify the inpainted regions in videos. In this paper, we present a Trusted Video Inpainting Localization network (TruVIL) with excellent robustness and generalization ability. Observing that high-frequency noise can effectively unveil the inpainted regions, we design deep attentive noise learning in multiple stages to capture the inpainting traces. Firstly, a multiscale noise extraction module based on 3D High Pass (HP3D) layers is used to create the noise modality from input RGB frames. Then the correlation between such two complementary modalities are explored by a cross-modality attentive fusion module to facilitate mutual feature learning. Lastly, spatial details are selectively enhanced by an attentive noise decoding module to boost the localization performance of the network. To prepare enough training samples, we also build a frame-level video object segmentation dataset (VOS2k5) with 2500 videos and pixel-level annotation for all frames. Both quantitative and qualitative evaluations on various inpainted videos verify the robustness against video compression and generalization ability of TruVIL.
- 5 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 22 (6) , 7215-7228
Trusted Video Inpainting Localization via Deep Attentive Noise Learning
Digital video inpainting technique has been substantially improved with deep learning in recent years. It may be used as malicious manipulation to remove important objects for creating forged videos. As such it is significant to blindly identify the inpainted regions in videos. In this paper, we present a Trusted Video Inpainting Localization network (TruVIL) with excellent robustness and generalization ability. Observing that high-frequency noise can effectively unveil the inpainted regions, we design deep attentive noise learning in multiple stages to capture the inpainting traces. Firstly, a multiscale noise extraction module based on 3D High Pass (HP3D) layers is used to create the noise modality from input RGB frames. Then the correlation between such two complementary modalities are explored by a cross-modality attentive fusion module to facilitate mutual feature learning. Lastly, spatial details are selectively enhanced by an attentive noise decoding module to boost the localization performance of the network. To prepare enough training samples, we also build a frame-level video object segmentation dataset (VOS2k5) with 2500 videos and pixel-level annotation for all frames. Both quantitative and qualitative evaluations on various inpainted videos verify the robustness against video compression and generalization ability of TruVIL.