Comparative Study of MTCNN and YuNet for Deepfake Detection

M. Tanmay Adithya, C. Tanush, N. Kathyaini, D. Mohit Reddy, G. Mary Swarna Latha

2025

Abstract

In recent years, deepfakes have become a prominent digital threat, raising concerns about the potential harm they can inflict on personal privacy and the fabric of society as trust in visual evidence becomes increasingly compromised. This paper provides an in-depth comparative analysis of MTCNN Vs YuNet face detection algorithms specifically focused on deepfake detection use cases. This work compares several baseline face detection models, systematically integrated into an InceptionResNetV1 model for classification to analyze which preprocessing technique yields the optimal performance for detecting facial manipulation techniques. To demonstrate the efficacy of this method, thorough experimental evaluations were conducted on the newly proposed OpenForensics dataset, which is characterized by diverse cases and rich face-level annotations, leveraging multiple faces in a single image. The results of each of the three reconfigurable system-level implementations are consistent in that the YuNet-based pipeline gives a significant improvement over the MTCNN-based system across all the core performance metrics (accuracy 57.2% vs 52.2%; precision 55.0% vs 51.6%; recall 82.8% vs 81.1%; and F1-score 66.1% vs 63.1%). Moreover, YuNet processes images much faster, at 0.008 seconds per image on average, compared to MTCNN's 0.024 seconds per image, indicating a 3x computational efficiency improvement. The YuNet pipeline also obtains a more accurate Area Under the ROC Curve score (0.624 vs 0.544), which measures the ability to accurately classify authentic and manipulated facial imagery across various classification thresholds. When analyzed more in-depth through confusion matrices, YuNet shows fewer false negatives as well, proving to be more effective at identifying deepfake images correctly. These findings collectively suggest that YuNet's enhanced detection capabilities, coupled with its architecture optimized for low-latency processing, make it significantly more suitable for real-time deepfake detection applications.

Paper Citation

in Harvard Style

Adithya M., Tanush C., Kathyaini N., Reddy D. and Latha G. (2025). Comparative Study of MTCNN and YuNet for Deepfake Detection. In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - Volume 1: ICRDICCT`25; ISBN 978-989-758-777-1, SciTePress, pages 615-624. DOI: 10.5220/0013870300004919

in Bibtex Style

@conference{icrdicct`2525,
author={M. Adithya and C. Tanush and N. Kathyaini and D. Reddy and G. Latha},
title={Comparative Study of MTCNN and YuNet for Deepfake Detection},
booktitle={Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - Volume 1: ICRDICCT`25},
year={2025},
pages={615-624},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013870300004919},
isbn={978-989-758-777-1},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - Volume 1: ICRDICCT`25
TI - Comparative Study of MTCNN and YuNet for Deepfake Detection
SN - 978-989-758-777-1
AU - Adithya M.
AU - Tanush C.
AU - Kathyaini N.
AU - Reddy D.
AU - Latha G.
PY - 2025
SP - 615
EP - 624
DO - 10.5220/0013870300004919
PB - SciTePress