Deepfake video detection: YOLO-Face convolution recurrent approach.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• The proliferation of hyper-realistic deepfake videos poses significant challenges to digital media authenticity.
• Detecting manipulated videos is crucial due to their potential for widespread negative societal impact.

Purpose of the Study:

• To introduce a novel deepfake video detection framework, YOLO-CRNNs.
• To enhance the accuracy and reliability of deepfake detection systems.

Main Methods:

• Utilized YOLO-Face detector for identifying face regions in video frames.
• Employed a fine-tuned EfficientNet-B5 for spatial feature extraction from detected faces.
• Integrated Bidirectional Long Short-Term Memory (Bi-LSTM) networks to extract temporal features from facial data sequences.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 89.35% on the c23 dataset.
• Obtained 89.38% accuracy, 83.15% recall, 85.55% precision, and 84.33% F1-measure for the pasting data approach.
• Demonstrated superior performance compared to existing state-of-the-art deepfake detection methods.

Conclusions:

• The proposed YOLO-CRNNs method offers a robust and effective solution for deepfake video detection.
• The framework's ability to extract both spatial and temporal features contributes to its high detection performance.
• This research provides a valuable advancement in combating the challenges posed by deepfake technology.