Exploiting Prediction Error Inconsistencies through LSTM-based Classifiers to Detect Deepfake Videos.
Irene AmeriniRoberto CaldelliPublished in: IH&MMSec (2020)
Keyphrases
- prediction error
- motion compensated
- video sequences
- motion vectors
- bit rate
- detection method
- decision trees
- inter frame
- video data
- knowledge base
- training set
- recurrent neural networks
- reversible watermarking
- power spectral density
- empirical risk
- video frames
- image quality
- training data
- linear prediction
- feature selection
- svm classifier
- motion compensation
- training samples
- feature vectors
- block matching
- majority vote
- motion compensated prediction
- three dimensional
- machine learning