30/05/2025
Research on Deep Learning-Based Robust Video Watermarking
Speaker: Jiaxuan Lin
Abstract: Against the backdrop of the rapid development of digital media technology, the creation, dissemination, and acquisition of video content have become unprecedentedly convenient, greatly enriching the public's audio-visual enjoyment. However, this convenience also poses severe challenges to copyright protection, especially with the surge in malicious video editing behaviors. Such actions not only infringe upon the intellectual property rights of original creators but also disrupt market order and damage the integrity and authenticity of video content. To address this challenge, this study proposes a robust video watermarking algorithm based on deep learning, aimed at protecting copyright owners from malicious video editing and preparing for anti-camcorder watermark technology that may be needed in the future.
30/05/2025
Various Dynamic Biometric Authentication
Speaker: Yi Zhao
Abstract: Dynamic biometric authentication represents a transformative approach to identity verification by leveraging unique behavioral and physiological patterns that evolve during user interaction. Unlike static biometrics (e.g., fingerprints, facial recognition), this technology analyzes real-time dynamic features such as keystroke dynamics, EEG, ECG, EMG, gait, vein patterns, gestures, and PPG. Key advantages include enhanced anti-spoofing capabilities—since dynamic traits are inherently difficult to replicate—and seamless user experiences through passive authentication. To further optimize the performance of dynamic biometric systems, our study introduces Compact Data Learning (CDL), a method designed to reduce the size and complexity of the training dataset while maintaining high recognition accuracy.
17/03/2025
Multi-media Forgery Detection and Localization via Higher-order Network
Speaker: Jiahao Huang
Abstract: With the rapid development of image editing and deepfake synthesis technologies, it has become increasingly easy to modify image, video, or audio using editing or generation techniques, allowing for the creation of false information. The misuse of these technologies poses a severe security challenge to society. Existing methods typically rely on manipulation traces, including artifacts left behind by editing processes and compression. However, these traces are often subtle and easily lost in detail. Therefore, this research proposal uses high-order network modeling to represent complex group relationships. We design multiple hypergraph neural network methods for multi-media forgery detection and localization.
27/02/2025
Research on Image Security in Quantum Technology Framework
Speaker: Zheng Xin
Abstract: This research focuses on the exploration of image security within the framework of quantum technology. Initially, it expounds on the core concepts of quantum technology, along with their unique advantages in information processing and computation. These quantum features are the cornerstone for the subsequent exploration of image-related security research. In the current stage study, the application of quantum technology to image processing security is investigated. The research mainly delves into implementing image copyright protection and encryption techniques in quantum systems. By developing innovative, quantum-based image processing algorithms, the security and integrity of image data are enhanced.
27/02/2025
Fine-Grained Understanding for Land Cover Change Information on High-Resolution Remote Sensing Images
Speaker: Junqing Huang
Abstract: Remote sensing (RS) plays a crucial role in optimizing the spatial configuration of terrestrial land resources. With the advancement of RS technology, the quality of high-resolution images (HRIs) has significantly improved. Change detection (CD) based on HRIs has become an important method for land resource surveys. However, understanding changes in RS image scenes is a complex systems engineering task, and there are still many critical technologies that have not been fully resolved or perfected. The two main challenges currently faced in CD are insufficient multi-scale feature fusion and inadequate redundant information processing.
09/01/2025
Computerized Respiratory Sound Analysis: From Deep Learning to Embedded Applications
Speaker: Fan Wang
Abstract: In the seminar, Fan Wang will introduce his recent work on respiratory sound classification. He will first discuss the basic characteristics and classification methods of respiratory sounds, and then introduce the application of deep learning to respiratory sound diagnosis. The presentation will also cover the implementation of lightweight deep learning models on embedded devices for real-time classification. Finally, we will discuss in detail the advantages and limitations of these methods in clinical and resource-constrained environments.
09/01/2025
Watermarking Technology For Artificial Intelligence Generated Content
Speaker: Qin Zhao
Abstract: With the rapid development of Artificial Intelligence Generated Content (AIGC), watermarking techniques have become essential for securing intellectual property rights and ensuring content traceability. However, existing watermarking methods face issues related to their vulnerability to tampering, limited data capacity, and dependency on model structure modification, especially within Latent Diffusion Models (LDM). To address these challenges, this presentation focuses on a novel, high-capacity watermarking approach specifically designed for LDM-based AIGC, leveraging a combination of robust embedding strategies and innovative training processes.
25/10/2024
Medical Data Security: Utilizing Watermarking and Cryptography
Speaker: Bowen Meng
Abstract: In the evolving landscape of smart healthcare systems, the secure transmission and integrity of Electronic Patient Records (EPR) are crucial. Digital watermarking is a reliable method for safeguarding digital images. However, ensuring the security and confidentiality of patient data in medical imaging remains a challenge. We have developed several hiding schemes for medical images. Methods combine fragile and robust watermarking techniques, along with cryptography. The image segmentation method is also used to select the Region of Non-Interest (RONI) for watermark embedding for maintaining losslessness in the Region of Interest (ROI), which is crucial for medical image. Experiments show the enhanced performance of the proposed method in terms of security, authenticity, and integrity, while preserving the high quality of medical images.
25/10/2024
Deep Learning Approaches to Forensic Detection
Speaker: Zhiyao Xie
Abstract: Digital images are evolved into an indispensable carrier for information dissemination in the prosperous multimedia era today. However, with the rise of increasingly user-friendly image editing software, the integrity and security of digital image information content are becoming increasingly vulnerable to manipulation and misuse. This creates significant challenges for digital forensics, making it harder to verify image authenticity and raising concerns about the trustworthiness of visual media. As a result, tamper detection has become a crucial research area, focused on developing reliable, efficient methods to distinguish between authentic and manipulated content, while adapting to the evolving landscape of digital manipulation.