Predicting manipulated regions in deepfake videos using convolutional vision transformers

Mohan  Bhandari; Sushant Shrestha; Utsab   Karki; Santosh  Adhikari; Rajan   Gaihre

doi:10.59400/cai.v2i2.1409

Predicting manipulated regions in deepfake videos using convolutional vision transformers

Mohan Bhandari Department of Science and Technology, Samriddhi College, Lokanthali, Bhaktapur 44800, Nepal
Sushant Shrestha Department of Computer Engineering, Kantipur Engineering College, Dhapakhel, Lalitpur 44700, Nepal
Utsab Karki Department of Computer Engineering, Kantipur Engineering College, Dhapakhel, Lalitpur 44700, Nepal
Santosh Adhikari Department of Computer Engineering, Kantipur Engineering College, Dhapakhel, Lalitpur 44700, Nepal
Rajan Gaihre Department of Computer Engineering, Kantipur Engineering College, Dhapakhel, Lalitpur 44700, Nepal

Ariticle ID: 1409

398 Views, 147 PDF Downloads

DOI: https://doi.org/10.59400/cai.v2i2.1409

Keywords: face detection; machine learning; vision transformer; convolution neural networks; Grad-CAM

Abstract

Deepfake technology, which uses artificial intelligence to create and manipulate realistic synthetic media, poses a serious threat to the trustworthiness and integrity of digital content. Deepfakes can be used to generate, swap, or modify faces in videos, altering the appearance, identity, or expression of individuals. This study presents an approach for deepfake detection, based on a convolutional vision transformer (CViT), a hybrid model that combines convolutional neural networks (CNNs) and vision transformers (ViTs). The proposed study uses a 20-layer CNN to extract learnable features from face images, and a ViT to classify them into real or fake categories. The study also employs MTCNN, a multi-task cascaded network, to detect and align faces in videos, improving the accuracy and efficiency of the face extraction process. The method is assessed using the FaceForensics++ dataset, which comprises 15,800 images sourced from 1600 videos. With an 80:10:10 split ratio, the experimental results show that the proposed method achieves an accuracy of 92.5% and an AUC of 0.91. We use Gradient-Weighted Class Activation Mapping (Grad-CAM) visualization that highlights distinctive image regions used for making a decision. The proposed method demonstrates a high capability of detecting and distinguishing between genuine and manipulated videos, contributing to the enhancement of media authenticity and security.

References

[1] Karnouskos S. Artificial Intelligence in Digital Media: The Era of Deepfakes. IEEE Transactions on Technology and Society. 2020; 1(3): 138-147. doi: 10.1109/tts.2020.3001312

[2] Grobler GD. Narrative strategies in the creation of animated poetry-film [PhD thesis]. University of South Africa; 2021.

[3] Wodajo D, Atnafu S, Akhtar Z. Deepfake video detection using generative convolutional vision transformer. Available online: https://arxiv.org/abs/2307.07036 (accessed on 20 May 2024).

[4] Heidari A, Jafari Navimipour N, Dag H, et al. Deepfake detection using deep learning methods: A systematic and comprehensive review. WIREs Data Mining and Knowledge Discovery. 2023; 14(2). doi: 10.1002/widm.1520

[5] Kearns L, Alam A, Allison J. Synthetic media authentication threats: Detection using a combination of neural network and blockchain technology. Available online: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4658121 (accessed on 20 May 2024).

[6] Chesney R, Citron DK. Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security. SSRN Electronic Journal. 2018. doi: 10.2139/ssrn.3213954

[7] Masood M, Nawaz M, Malik KM, et al. Deepfakes generation and detection: state-of-the-art, open challenges, countermeasures, and way forward. Applied Intelligence. 2022; 53(4): 3974-4026. doi: 10.1007/s10489-022-03766-z

[8] Montserrat DM, Hao H, Yarlagadda SK, et al. Deepfakes Detection with Automatic Face Weighting. In: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW); 2020. doi: 10.1109/cvprw50498.2020.00342

[9] Afchar D, Nozick V, Yamagishi J, et al. MesoNet: a Compact Facial Video Forgery Detection Network. In: 2018 IEEE International Workshop on Information Forensics and Security (WIFS); 2018. doi: 10.1109/wifs.2018.8630761

[10] Ha H, Kim M, Han S, et al. Robust Deep Fake Detection Method based on Ensemble of ViT and CNN. In: Proceedings of the 38th ACM/SIGAPP Symposium on Applied Computing; 2023. doi: 10.1145/3555776.3577769

[11] Hasan FS. FaceForensics-1600 videos-preprocess. Available online: https://www.kaggle.com/datasets/farhansharukhhasan/faceforensics1600-videospreprocess?rvi=1 (accessed on 23 May 2024).

[12] Jose EMG, Haridas MTP, Supriya MH. Face Recognition based Surveillance System Using FaceNet and MTCNN on Jetson TX2. 2019 5th International Conference on Advanced Computing & Communication Systems (ICACCS). Published online March 2019. doi: 10.1109/icaccs.2019.8728466

[13] Selvaraju RR, Cogswell M, Das A, et al. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization. In: 2017 IEEE International Conference on Computer Vision (ICCV); 2017; Venice, Italy. pp. 618-626. doi: 10.1109/iccv.2017.74

Published

2024-07-01

How to Cite

Bhandari, M., Shrestha, S., Karki, U., Adhikari, S., & Gaihre, R. (2024). Predicting manipulated regions in deepfake videos using convolutional vision transformers. Computing and Artificial Intelligence, 2(2), 1409. https://doi.org/10.59400/cai.v2i2.1409

Download Citation

Issue

Vol. 2 No. 2 (2024)

Section

Article

This work is licensed under a Creative Commons Attribution 4.0 International License.

Editor-in-Chief

Prof. Shaohua Wan
University of Electronic Science and Technology of China, China

eISSN

3029-2786

Publication Frequency

Semi-annual

About the Publisher

Academic Publishing insists on taking academic exchange and publication as the main line, carrying out comprehensive management based on science and technology, and fully exploring excellent international publishing resources. Within 5 years, it will form a strategic framework and scale with science (S), technology (T), medicine (M), education (E), and humanities and arts (H) as the main publishing fields. Academic Publishing is headquartered in Singapore and based in Malaysia, with the United States and China providing the main scientific and academic resources. At the same time, it has established long-term good cooperative relations with other publishing companies, scientific research communities, and academic organizations in more than a dozen countries and regions. Academic Publishing uses English and Chinese as its main publishing languages, mainly publishing books, journals, and conference papers in print and online. The vast majority of publications follow the international open access policy, providing stable and long-term quality and professional publications. With the joint efforts of the expert team and our professional editorial team, our publications will gradually be indexed by international databases in stages to provide convenient and professional retrieval for various scholars. At the same time, manuscripts we accept will be subject to the peer review principle, and cutting-edge and innovative research articles will be preferentially accepted for peer reference and discussion. All kinds of our publications are welcome for peer to contribute, access, and download.

more

Volume Arrangement

2024

Vol 2, No 2 (2024)

Vol 2, No 1 (2024)

2023

Vol 1, No 1 (2023)

Featured Articles

Identifying voices using convolution neural network models AlexNet and ResNet

Deep learning (DL) techniques which implement deep neural networks became popular due to the increase of high-performance computing facilities. DL achieves higher power and flexibility due to its ability to process many features when it deals with unstructured data. DL algorithm passes the data through several layers; each layer is capable of extracting features progressively and passes it to the next layer. Initial layers extract low-level features, and succeeding layers combine features to form a complete representation. This research attempts to utilize DL techniques for identifying sounds. The development in DL models has extensively covered classification and verification of objects through images. However, there have not been any notable findings concerning identification and verification of the voice of an individual from different other individuals using DL techniques. Hence, the proposed research aims to develop DL techniques capable of isolating the voice of an individual from a group of other sounds and classify them based on the use of convolutional neural networks models AlexNet and ResNet, that are used in voice identification. We achieved the classification accuracy of ResNet and AlexNet model for the problem of voice identification is 97.2039 % and 65.95% respectively, in which ResNet model achieves the best result.

Revolutionizing Neurosurgery and Neurology: The transformative impact of artificial intelligence in healthcare

The integration of artificial intelligence (AI) has brought about a paradigm shift in the landscape of Neurosurgery and Neurology, revolutionizing various facets of healthcare. This article meticulously explores seven pivotal dimensions where AI has made a substantial impact, reshaping the contours of patient care, diagnostics, and treatment modalities. AI’s exceptional precision in deciphering intricate medical imaging data expedites accurate diagnoses of neurological conditions. Harnessing patient-specific data and genetic information, AI facilitates the formulation of highly personalized treatment plans, promising more efficacious therapeutic interventions. The deployment of AI-powered robotic systems in neurosurgical procedures not only ensures surgical precision but also introduces remote capabilities, mitigating the potential for human error. Machine learning models, a core component of AI, play a crucial role in predicting disease progression, optimizing resource allocation, and elevating the overall quality of patient care. Wearable devices integrated with AI provide continuous monitoring of neurological parameters, empowering early intervention strategies for chronic conditions. AI’s prowess extends to drug discovery by scrutinizing extensive datasets, offering the prospect of groundbreaking therapies for neurological disorders. The realm of patient engagement witnesses a transformative impact through AI-driven chatbots and virtual assistants, fostering increased adherence to treatment plans. Looking ahead, the horizon of AI in Neurosurgery and Neurology holds promises of heightened personalization, augmented decision-making, early intervention, and the emergence of innovative treatment modalities. This narrative is one of optimism and collaboration, depicting a synergistic partnership between AI and healthcare professionals to propel the field forward and significantly enhance the lives of individuals grappling with neurological challenges. This article provides an encompassing view of AI’s transformative influence in Neurosurgery and Neurology, highlighting its potential to redefine the landscape of patient care and outcomes.

Enhancing user experience in large language models through human-centered design: Integrating theoretical insights with an experimental study to meet diverse software learning needs with a single document knowledge base

This paper begins with a theoretical exploration of the rise of large language models (LLMs) in Human-Computer Interaction (HCI), their impact on user experience (HX) and related challenges. It then discusses the benefits of Human-Centered Design (HCD) principles and the possibility of their application within LLMs, subsequently deriving six specific HCD guidelines for LLMs. Following this, a preliminary experiment is presented as an example to demonstrate how HCD principles can be employed to enhance user experience within GPT by using a single document input to GPT’s Knowledge base as new knowledge resource to control the interactions between GPT and users, aiming to meet the diverse needs of hypothetical software learners as much as possible. The experimental results demonstrate the effect of different elements’ forms and organizational methods in the document, as well as GPT’s relevant configurations, on the interaction effectiveness between GPT and software learners. A series of trials are conducted to explore better methods to realize text and image displaying, and jump action. Two template documents are compared in the aspects of the performances of the four interaction modes. Through continuous optimization, an improved version of the document was obtained to serve as a template for future use and research.

Clustering data analytics of urban land use for change detection

In this study, the author proposes and details a workflow for the spatial-temporal demarcation of urban areal features in 8 cities of Tamilnadu, India. During the inception phase, functional requirements and non-functional parameters are analyzed and designed, within a suitable pixel area and object-oriented derived paradigm. Land use categories are defined from OpenStreetMap (OSM) related works with the scope of conducting climate change, using multispectral sensors onboard Landsat series. Furthermore, we augment the bands dataset with Spatially Invariant Feature Transform (SIFT), Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-Up Index (NDBI), Leaf Area Index (LAI), and Texture base indices, as a means of spatially integrating auto-covariance to stationarity patterns. In doing so, change detection can be pursuit by scaling up the segmentation of regional/zonal boundaries in a multi-dimensional environment, with the aid of Wide Area Networks (WAN) cluster computers such as the BEOWULF/Google Earth Engine clusters. GeoAnalytical measures are analyzed in the design of local and zonal spatial models (GRID, RASTER, DEM, IMAGE COLLECTION). Finally, multi variate geostatistical works are ensued for precision and recall in predictive data analytics. The author proposes reusing machine learning tools (filtering by attribute-based indexing in PaaS clouds) for pattern recognition and visualization of features and feature collection.

Application of computer vision in livestock and crop production—A review

Nowadays, it is a challenge for farmers to produce healthier food for the world population and save land resources. Recently, the integration of computer vision technology in field and crop production ushered in a new era of innovation and efficiency. Computer vision, a subfield of artificial intelligence, leverages image and video analysis to extract meaningful information from visual data. In agriculture, this technology is being utilized for tasks ranging from disease detection and yield prediction to animal health monitoring and quality control. By employing various imaging techniques, such as drones, satellites, and specialized cameras, computer vision systems are able to assess the health and growth of crops and livestock with unprecedented accuracy. The review is divided into two parts: Livestock and Crop Production giving the overview of the application of computer vision applications within agriculture, highlighting its role in optimizing farming practices and enhancing agricultural productivity.