multimedia-forensics

This repo contains the ENF-WHU audio recording dataset collected around Wuhan University campus and the MATLAB programs for electronic network frequency (ENF) detection, enhancement, and robust estimation, in ENF-based audio forensic applications.

Official code for FAME: A lightweight spatio-temporal framework for Deepfake Model Attribution using VGG-based feature extraction and temporal attention. Outperforms prior methods across DFDM, FF++, and FakeAVCeleb datasets.

Includes three methods for median filtering forensics in digital images

MATLAB code for the Robust LAD-LRT ENF Detection (ICASSP 2023)

Structure matters: analyzing videos via graph neural networks for social media platform attribution

Official implementation of the paper ``DeepFeatureX-SN: Generalization of deepfake detection via contrastive learning'', accepted for publication at the Springer journal 'Multimedia Tools and Application'.

Official code for CapST (ACM TOMM 2025): A Capsule Network and Temporal Attention-based framework for Deepfake Model Attribution. Achieves efficient and accurate source identification across DFDM and GAN-generated datasets.

Official PyTorch implementation of "Depthtective: A Depth-Aware Framework for Spatio-Temporal Deepfake Detection" (Array, 2026). Detects manipulated videos from the RGB and depth residuals of just two adjacent frames.

Closed-Set SDI Using log-Mel Spectrograms from Videos (IEEE Access)

A unified multi-layer verification engine that detects real human presence and validates physical, biological, and temporal consistency in digital media to produce an explainable trust score

ClipCert: Web-based video authenticity checker. Compare two videos frame-by-frame using advanced similarity algorithms (pHash/pixel comparison). Secure, client-side processing ensures data never leaves your browser.

Contribution: This repository extends the original CLIP implementation by adding training code for an SVM classifier using CLIP-extracted features, with detailed setup instructions in README_TRAIN.md.

High-throughput video frame extraction and multimedia forensic artifact pipeline in Rust.

Verify the authenticity of web content with a lightweight, browser-first HTML utility.