Deepfake Detector
Recent advancements in machine learning, deep learning and computer vision made it extremely easy for anyone to create fake videos, images and audios of anyone. Using deep learning techniques like Autoencoders and Generative Adversarial Networks (GANs)
2025-06-28 16:26:05 - Adil Khan
Deepfake Detector
Project Area of Specialization Software EngineeringProject SummaryRecent advancements in machine learning, deep learning and computer vision made it extremely easy for anyone to create fake videos, images and audios of anyone. Using deep learning techniques like Autoencoders and Generative Adversarial Networks (GANs), one can manipulate given video or an image of target person. This technique helps in spreading disinformation on social media platforms like Twitter, Facebook, YouTube etc. In past few years deepfakes of politicians, actors and actresses, celebrities have gone viral, which caused a fear of security and privacy between individuals. Sometimes, these deepfakes seems to be entertaining but at the same time it is easily weaponized to disgrace some personality. The first misuse of this technology was creation of Revenge Porn, which poses a serious threat to women. Some of the women community simply using this technology to attract public for so called fame. Fraudulent activities and disrupting government functioning is also caused by Deepfake technology.
Most of the people today are confused to classify video as real or fake. In Social Media context, they developed their own systems to classify uploaded digital content as deepfake or not. As Twitter has taken measures to check synthesized data to stop disinformation among public. Facebook is also an active member of Deepfake Detection Challenge and working to detect deepfakes.
Deepfakes are the deadly tools of fifth generation warfare. Therefore, it is very important to solve this problem by developing an automated deep forgery detector method that can reliably detect fake videos. Deepfakes can be generated using different methods, namely face swap, lip sync, puppet master, etc. We aim to propose a framework based on hybrid integration, which is lightweight compared with CNN integration, and has better generalization compared with other CNN-based technologies and convolution classifiers like SVM, KNN etc. We will use combination of Vision Transformer with either of CNN architecture to develop a robust model which can classify as Video "Real" or "Fake.
Project ObjectivesThe aims and objectives of the proposed work are:
- To develop an effective end-to-end deep learning method that can reliably be used to detect the deepfakes videos and images.
- To develop an interactive deepfakes detector app that can take the input from the user and automatically detect whether the given video/image is bonafide or fake.
The proposed method will be implemented using the deep learning. Vision Transformers, competitive alternative to CNN which are currently State of The Art (SOTA) in image classification, are used in combination with a CNN architecture. There are many datasets available for deepfake detection, like FaceForensics++, DFDC, DFDC Preview, WLDR, etc. We used FaceForensics++ with WLDR to develop a model based on ViT and CNN.
First of all, the MTCNN will be used to detect faces in a video, and then we will extract those frames from the video containing those faces. These extracted frames will then be passed to Efficient-Net B7, which will work as a feature extractor. Those processed images or features will then be further processed by Vision Transformer. The extracted features are very low level detailed, and these will simplify the training process of Vision Transformer.
Benefits of the ProjectThe deepfake detector will be a beneficial product for different kind of situations. During crises someone may create a deepfake video of any influential person and can say some stupid things which the actual person never said. In these kind of situation deepfake detector will be extremely helpful to evaluate these kinds of videos whether these are real or fake. Deepfake detector will also be extremely helpful in those situations where dirty minds create fake porn video of any respected person and try to disrupt his image in front of society. The deepfake detector will then be used to detect the forgeries and clean that person name.
Technical Details of Final DeliverableThe final deliverables will be web-based application using React which will be used to detect any kind of deepfake video.
The user will be able to upload the video through the user interface. This video will then be stored on the local storage of the user and model will perform its task on the backend. At the end, results will be shown on the output screen classifying a video as deepfake or not. The deepfake detector will be very robust and reliable. Another aspect will be the research paper based on the research work done to detect forged videos.
Final Deliverable of the Project Software SystemCore Industry ITOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 80000 | |||
| LCD Display | Equipment | 1 | 35000 | 35000 |
| NVIDIA GPU | Equipment | 1 | 35000 | 35000 |
| Cloud and Hosting Services | Miscellaneous | 1 | 10000 | 10000 |