计算机科学
认证(法律)
一致性(知识库)
水准点(测量)
协议(科学)
人工智能
数字水印
推论
数据挖掘
数字签名
机器学习
数字媒体
社会化媒体
麻省理工许可证
加权
面子(社会学概念)
数字取证
特征提取
秩(图论)
深度学习
计算机安全
统一建模语言
方向(向量空间)
入侵检测系统
数据科学
作者
Aareen Dholakia,Ananth Balvalli,Austin Paul,Shruti Dalvi,Shubhangi L. Vaikole
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-12-20
卷期号:: 721-726
标识
DOI:10.1109/cicn67655.2025.11367828
摘要
The increase in the usage of Deepfake Technology poses both known and unknown threats to digital media integrity and authenticity, which has led to the development of robust detection techniques across multiple media sectors. To tackle this alarming risk posed by manipulated digital media, this paper presents a novel multi-modal Deepfake Detection system that integrates image, video, and audio analysis with blockchain-based verification for tamper-proof authentication. Our model uses advanced deep learning tools and frameworks including BlazeFace for face detection, EfficientNetB4Att for image analysis, and Tortoise for audio processing. On the FaceForensics++ dataset, it achieves an excellent score with 93.6% accuracy. The decision-level fusion mechanism with adaptive weighting allows for accurate detection. It provides accurate results for partial inputs, which are verified through the immutable and tamperproof report generated via blockchain technology. Experimental validation across benchmark datasets demonstrates significant improvements over existing single-modality approaches, with real-time inference capabilities of 25ms per analysis. The system identifies the main weaknesses of the current deepfake detection methods and strengthens them by providing a complete, scalable, and safe media authentication tool that can be used on multiple social media platforms, blogs, and legal applications. The efficiency and consistency of this Deepfake Detection System in precisely discriminating between real and distorted media across varied scenarios and circumstances have been validated through meticulous testing and evaluation.
科研通智能强力驱动
Strongly Powered by AbleSci AI