International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 10 Issue: 08 | Aug 2023
p-ISSN: 2395-0072
www.irjet.net
A Neural Network Approach to Deep-Fake Video Detection Prajwal Chunarkar1, Vivek Upadhyay1, Sagar Sanap1 , Anirudh Talmale1, Prof. Varshapriya J N 2 1BTech Student, Dept of Computer Engineering and IT, VJTI College, Mumbai, Maharashtra, India
2Associate Professor, Dept of Computer Engineering and IT, VJTI College, Mumbai, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Deep learning has been used to solve a variety of
develop technology that can identify fakes, so that the DF can be spotted and prevented from spreading over the internet.
complex challenges, including big data analytics, computer vision, and human-level control. Deep learning developments, on the other hand, have been used to develop applications that pose a risk to anonymity, democracy, and national security. Deepfake is a recent example of a deep learning-powered technology. Deepfake algorithms can generate fake photographs and videos that are difficult to tell apart from real ones. As a result, the development of technology that can automatically identify and measure the integrity of digital visual media is critical.Using artificially intelligent software to build the DF is an easy process. However, detecting these DF poses a significant challenge. Since it is difficult to train the algorithm to detect the DF. Using Convolutional Neural Networks and Recurrent Neural Networks, we have made progress in detecting the DF. At the frame stage, the system employs a convolutional Neural network (CNN) to extract features. These characteristics are used to train a recurrent neural network (RNN), which learns to classify whether or not a video has been manipulated and can detect frame temporal inconsistencies caused by DF creation tools. Expected outcome as compared to a large number of fake videos gathered from a regular data set. We demonstrate how using a simple architecture, our system can achieve competitive results in this mission.
2. LITERATURE REVIEW Deep fake video's exponential development and illicit usage pose a serious threat to government, justice, and public trust. As a result, the market for fake video review, monitoring, and interference has risen. The following are some of the relevant studies in deep fake identification. Exposing AI Created Fake Videos by Detecting Eye Blinking [1] explains a new approach for exposing fake face videos produced by deep neural network models. The approach relies on the identification of eye blinking in videos, which is a physiological signal that isn't well shown in the fake videos. The method is tested on eye-blinking recognition datasets and shows positive results when it comes to detecting videos created with Deep Neural Network based software DF. Their system relies solely on the absence of blinking as an identification clue. However, other factors such as teeth enchantment, lines on the forehead, and so on must be noticed when detecting a deep fake. All of these criteria are taken into account by our system.
Key Words: Deepfake Video Detection, convolutional, Neural network (CNN), recurrent neural network (RNN)
Using capsule networks to detect forged images and videos [2] uses a method that uses a capsule network to detect forged, manipulated images and videos in different scenarios, like replay attack detection and computergenerated video detection.
1.INTRODUCTION Recent advances in artificial intelligence (AI) and cloud computing technology have resulted in rapid developments in audio, video, and image processing techniques. Deepfakes are the term for this kind of fake media material AI-based tools will also exploit media in more convincing ways, such as duplicating a public figure's voice or superimposing one person's face over another's body. Deep generative adversarial models that can exploit video and audio clips generate "DeepFake." The spread of the DF through social media channels has become very popular, resulting in spamming and the dissemination of incorrect information. This form of DF would be bad, and will threaten and confuse ordinary citizens.
In their method, they have used random noise in the training phase which is not a good option. Still the model performed beneficial in their dataset but may fail on real time data due to noise in training. Our method is proposed to be trained on noiseless and real time datasets. Luo et al. [3] introduces JPEG error analysis to determine when bitmap images have been compressed previously, approximate quantization steps, and detect the quantization table in a JPEG image. Their ability to detect JPEG blocks as small as 8x8 assists in the identification of tampered areas in an image. Bianchi et al.
The importance of DF identification in such a case cannot be overstated. As a result, we present a new deep learningbased method for distinguishing between AI-generated fake videos (DF Videos) and real videos. It’s very important to
© 2023, IRJET
|
Impact Factor value: 8.226
J. H. Bappy et al. [4] introduce a method for manipulation localization that uses an LSTM network and an encoder-decoder system to construct a mapping from low resolution activations to pixel-wise predictions. CNNs have
|
ISO 9001:2008 Certified Journal
|
Page 675