Real-Fake Face Detection Based on Joint Multi-Layer CNN Structure and Data Augmentation

Weiting Bian

2023

Abstract

Nowadays, with the advancement of technology, various image editing and image generation tools have emerged, leading to the generation of fake face images. This has caused many issues, such as fraud and false information. Therefore, it is highly meaningful to use more effective methods to identify real and fake faces. The topic of this study is real-false face recognition grounded on the Convolutional Neural Network (CNN) model. CNN structure is utilized, consisting of data augmentation, resize, scale, convolution, pooling, and fully-connected layers (FC). Initially, both training and validation losses are relatively high for the training results, but as training iterations progress, the losses gradually decrease. Meanwhile, the accuracy of the model gradually improves after several rounds of iterations, ultimately reaching 90% on the training and validation sets. After being evaluated on an independent test dataset, the model achieved a 15.90% loss with a 93.63% accuracy. The model achieves high accuracy in predicting real and fake faces, demonstrating good performance and practicality. Lastly, effective recognition of real and fake faces can help people identify false information, avoiding panic, financial losses, and rumors to some extent. It plays a significant role in social stability.

Paper Citation

in Harvard Style

Bian W. (2023). Real-Fake Face Detection Based on Joint Multi-Layer CNN Structure and Data Augmentation. In Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML; ISBN 978-989-758-705-4, SciTePress, pages 25-29. DOI: 10.5220/0012800300003885

in Bibtex Style

@conference{daml23,
author={Weiting Bian},
title={Real-Fake Face Detection Based on Joint Multi-Layer CNN Structure and Data Augmentation},
booktitle={Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML},
year={2023},
pages={25-29},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012800300003885},
isbn={978-989-758-705-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML
TI - Real-Fake Face Detection Based on Joint Multi-Layer CNN Structure and Data Augmentation
SN - 978-989-758-705-4
AU - Bian W.
PY - 2023
SP - 25
EP - 29
DO - 10.5220/0012800300003885
PB - SciTePress