Deepfake Detection in Videos Using Long Short-Term Memory and CNN ResNext

Abstract

Deep-fake in videos is a video synthesis technique by changing the people’s face in the video with others’ face. Deep-fake technology in videos has been used to manipulate information, therefore it is necessary to detect deep-fakes in videos. This paper aimed to detect deep-fakes in videos using the ResNext Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) algorithms. The video data was divided into 4 types, namely video with 10 frames, 20 frames, 40 frames and 60 frames. Furthermore, face detection was used to crop the image to 100 x 100 pixels and then the pictures were processed using ResNext CNN and LSTM. The confusion matrix was employed to measure the performance of the ResNext CNN-LSTM algorithm. The indicators used were accuracy, precision, and recall. The results of data classification showed that the highest accuracy value was 90% for data with 40 and 60 frames. While data with 10 frames had the lowest accuracy with 52% only. ResNext CNN-LSTM was able to detect deep-fakes in videos well even though the size of the image was small.

Keywords

• deepfake
• resnext cnn
• lstm