Improving Video Deepfake Detection: A DCT-Based Approach with Patch-Level Analysis

View on arXiv ← Back to list

Authors: Luca Guarnera, Salvatore Manganello, Sebastiano Battiato

Published: 2023-10-17 12:30:46+00:00

AI Summary

This paper introduces a novel deepfake detection algorithm using Discrete Cosine Transform (DCT) analysis on video I-frames. By extracting beta components from AC coefficients of various facial patches (eyes, mouth, etc.), the algorithm identifies discriminative features for distinguishing real from deepfake videos.

Abstract

A new algorithm for the detection of deepfakes in digital videos is presented. The I-frames were extracted in order to provide faster computation and analysis than approaches described in the literature. To identify the discriminating regions within individual video frames, the entire frame, background, face, eyes, nose, mouth, and face frame were analyzed separately. From the Discrete Cosine Transform (DCT), the Beta components were extracted from the AC coefficients and used as input to standard classifiers. Experimental results show that the eye and mouth regions are those most discriminative and able to determine the nature of the video under analysis.

Key findings
The eye and mouth regions proved most discriminative. A Random Forest classifier achieved the best performance with an AUC of 94.55% on the eye region, slightly outperforming some state-of-the-art deep learning methods. The proposed method is computationally efficient and interpretable.
Approach
The approach extracts I-frames from videos. It then performs DCT on 8x8 blocks within different facial regions (eyes, mouth, etc.). Beta components from AC coefficients are used as features for standard classifiers (k-NN, SVM, Random Forest, etc.) to differentiate real and deepfake videos.
Datasets
FaceForensics++ and Celeb-DF (v2)
Model(s)
k-NN, Linear Discriminative Analysis (LDA), Support Vector Machine (SVM), Random Forest, Decision Tree, Gradient Boosting (GBoost)
Author countries
Italy
← Previous
Next →