A Note on Deepfake Detection with Low-Resources

Authors: Piotr Kawa, Piotr Syga

Published: 2020-06-09 11:07:08+00:00

AI Summary

This paper explores two methods for low-resource deepfake detection in videos. It enhances the MesoNet architecture by replacing original activation functions with novel ones, including the newly introduced 'Pish' function, leading to improved accuracy and consistency. Additionally, the authors present a preliminary deepfake detection method based on Local Feature Descriptors (LFD) that operates without significant computational power, achieving an Equal Error Rate (EER) of 0.28.

Abstract

Deepfakes are videos that include changes, quite often substituting face of a portrayed individual with a different face using neural networks. Even though the technology gained its popularity as a carrier of jokes and parodies it raises a serious threat to ones security - via biometric impersonation or besmearing. In this paper we present two methods that allow detecting Deepfakes for a user without significant computational power. In particular, we enhance MesoNet by replacing the original activation functions allowing a nearly 1% improvement as well as increasing the consistency of the results. Moreover, we introduced and verified a new activation function - Pish that at the cost of slight time overhead allows even higher consistency. Additionally, we present a preliminary results of Deepfake detection method based on Local Feature Descriptors (LFD), that allows setting up the system even faster and without resorting to GPU computation. Our method achieved Equal Error Rate of 0.28, with both accuracy and recall exceeding 0.7.

Key findings

Enhancing MesoNet with the Pish + Mish activation function combination achieved a test accuracy of 0.8689, slightly outperforming the baseline ReLU+Leaky ReLU (0.8676) and demonstrating the competitive performance of the novel Pish function. The LFD-based method, utilizing BRISK features and an SVM classifier, achieved an EER of 0.28 with accuracy and recall exceeding 0.7, offering a faster, GPU-free solution.

Approach

The authors enhance the MesoNet convolutional neural network by investigating and replacing its activation functions with newer, more effective ones (including their proposed 'Pish' function). Separately, they develop a preliminary deepfake detection method that uses Local Feature Descriptors (LFDs) like ORB and BRISK to extract features from video frames, which are then classified by a Support Vector Machine (SVM).

Datasets

Afchar et al.'s self-collected dataset (for MesoNet experiments), FaceForensics++ (subset for LFD experiments), Fashion MNIST, CIFAR-10 (for general activation function benchmarking).

Model(s)

MesoNet (specifically MesoInception-4), Support Vector Machine (SVM), SqueezeNet, DenseNet, EfficientNet.

Author countries

Poland

← Previous