Hello everyone welcome to IEEE expert. So today we are going to see project about very much important project for current society. So please go through the complete video.
This project very much helpful for the future generation. We are facing huge trouble in current days. Okay this project will be a solution for those issues. Okay please follow the video carefully.
So this is the project which you are going to do today. Deep fake face detection using conversion and neural net. So this is the project which you are going to implement. So this is the project base paper which is published on IEEE.
As well as we have considered two papers. These two papers completely belongs to recent year. IEEE 2023 recent year.
Deep fake face detection. So this is the project which you are going to do. Okay. You know, nowadays, okay.
Past one day before you know very well, Rashmiya Mandana deep fake face video went viral. Okay, so everyone initially thought that that belongs to Rashmiya Mandana. Later only they come to realize that it's that video completely made up of AI.
So nowadays AI ground very much significantly. That may be useful for various domains as well as various infrastructure. But we may land up in the huge trouble in the negative side of the AI also. So the video completely made by AA. So you know everyone thought that that video was real.
AA was that much accurate nowadays. Okay. So even Indian Union Minister warned that future generation of AA was very much scary. Okay. The statement not belongs to the common citizen.
The AA is scary. The statement revealed by Union Minister of India. So that much scary incident happened. Okay, so it's our important duty to overcome these issues using AI. Okay.
So also you know that every month hundreds of girls victim affected by the AI deep fake face videos major those girls are land against suicide thought. Okay, so it's not important. It's not good for your society, you know. So we have to prevent some method for the suicides, the fake face videos, deep fake face videos. We have to prevent some methodologies.
We have to implement some methodologies. So this project will be a solution for us. Okay. Even our Rajniya Mandana also said that the AI algorithm was completely scary for me.
I never thought that the AI will make this much dangerous like that. She also revealed in a revelation Instagram post. Okay.
So not only India. We are facing this issue even US also released a statement. We are completely watching the AI technology. We are completely analyzing the AI technology for deep fake face detection. Even US also released a statement.
So that much scare incident happened nowadays. So we have to adapt this technology. So completely you can't omit this AI.
So AI belongs to major advantages also if having two side of parts one is white another one is black okay we completely we cannot omit due to its black part dark side of the AI we have to take its positive also so what we are going to do means by using that same AI only we have to prevent that negative side of the AI so what we are going to do means we have to use some intelligence algorithm those algorithms can be capable of detecting the face fake videos that is the thing which we are going to See today, okay, so this is the project base paper. See that this project base paper they have used conversion neural network Without wasting much for the time. Let's go to the project presentation.
Let's see the project methodology So we are going to make deep fake face detection using artificial integers machine learning this is the project which we are going to discuss today so the project abstract was we are given that in the project abstract we have mentioned that we are going to take lstm long short time memory as a proposed algorithm okay in the basis paper they have taken cnn algorithm conversion neural network but it's mainly these two major drawbacks which we seen in the later part of the presentation cnn having huge drawbacks to work on this drawbacks we are going to make this deep wake phase detection using lstm long short time memory so this is the project abstract okay so first of all you have to know that what's the meaning of the fake face detection okay what's the meaning of deep wake space the deep fake faces are generated by the artificial intelligence algorithm completely as well as seamlessly okay it It can be able to replace the original face with the manipulated audio as well as manipulated video as well as it can able to blur some part of the face. It can able to hide the truth of the video. That is completely called as Deep Fake Face Detection.
So it's used for the tip purpose malicious contacts as well as the name spoiling purpose. It will leads to various problems. To work on that only we are going to implement the project.
First of all, you have to see what's the existing system of the product. The existing system means we have to mention that existing system completely belongs to the base paper. We have taken 2023 base paper. In the base paper, they have created this system with CNN, Convolutional Neural Network. But the major drawback was Convolutional Neural Network having four drawbacks.
We are going to explain that. The major drawback was limited temporal understanding. Convolutional Neural Network cannot be able to...
understand all the features present in the fake video as well as original video it cannot be able to distinguish these two features okay so it may it won't have it don't handle all the data optimally so which will lead to huge data loss that is a major drawback and large computational requirement to run this project to run the cnn architecture you need graphics processor system huge resources okay huge memory huge processor, graphical processor, gaming pc like that you need to run this project. It won't suitable for all the person you know. That is the major drawback. The third drawback was vulnerability to the adversarial network.
You can hack the cnn network. That is the huge drawback. Cnn network can be hacked by the adversarial attacks. So someone may hack this algorithm they can able to perform fake video as a real video they can able to hack this complete algorithm they can say the fake video into original video so that also lead to the accuracy training data imbalance what the meaning of training data imbalance means if you are giving less number of data to train means cl won't perform seamlessly it will lead to huge data loss to overcome that only we are going for the proposed system this is the project proposed system here we are going to make this complete project using Long short term memory LSTM this project completely going to make with a LSTM long short term memory so this is the major advantage here this project completely belongs to the AI algorithms okay it's a very robust process it's crucial to detect deep wake phase incident in continuous manner this algorithm will be helpful for that so in proposed system we are going to consider R and also recurrent neural network along with LSTM for making this system some more additional benefit okay it will give a little bit of more accuracy little bit of more accuracy for that purpose only we are going to use RNN here so then this is the project architecture as we mentioned earlier you have to take the data set in the data set we are going to collect various deep wake phase datas from the Kaggle those data this a data set we have to split into the three part training part testing part as well as validation part So we are going to take this dataset as a video format only. Those videos are always converted into frames.
Those frames are going to be trained. For the training only we are using LSTM here. What LSTM will do means, First of all, you have to take the dataset, you know those videos.
We have to apply pre-processing. In the pre-processing stage, we have to detect the faces. We have to extract the background from that. We have to NF the face. From that we have to crop the facial part only those facial features are trained under lsdm so lsdm will train all this data in the detection part whether you have to give any type of video or or else any type of fakes those uploaded video again extracted into multiple frames those frames part that comes under the pre-crossing part in training at testing stages also so those from the frames detector phase again come to the alex net for testing purpose then it will compare with the already trained values it will give results as a fake or real But the accuracy was huge here.
So this is the architecture. So these are the project models. We have taken four modules.
First module belongs to dataset collection. We are going to collect data under. Second module data pre-processing module. Pre-processing module you have to extract frames, faces, background extraction, noise removal under.
In the third module as we mentioned earlier it comes under training module. LSTM will split the data into training as well as validation model. It will fine tune all this value for the lstm network it will store all the features inside the lstm network then final evaluation testing model you have to give any type of video or whatever video you have moving for testing purpose it will extract the frame pre-process again it will compare with the already trained values it will give the result as fake video or original so okay this this is what the mod is so what are the software and the hardware required for this million this project means you need minimum i3 processor enough 4gb of ram enough talk for this project is available already on low end system itself for this project we need windows operating system as well as mac also supporting okay so coding language we are going to use python as a major language for website purpose we are using html css javascripts also okay so website framework this project using flash for the website major advantage of this project was uh overall efficiency of the cnn was uh sorry lst mask huge as well as it can able to handle 60 fps video 40 fps video 120 fps video seamlessly enhanced accuracy this gives 98 accuracy in real time okay those are the major advantage here so the conclusion was we are going to apply lstm by using lstm we are implementing this project for higher efficiency as well as we are permitting the complex structure of the cnn for the better understanding we are using lstm here so those are the reference project reference okay this is the project ppt explanation part uh be in touch with us again we are moving moving to the demo part okay next we are going for the demo plot let's see the project demonstration okay thank you let's move on to the project demo part to run this project we need anaconda navigator in the anaconda navigator we have created project environment while purchasing the project you will get the complete software links how to install how to execute complete support you can get so i'm just opening the project terminal uh let's move on to the project coding folder so this is the project coding folder we have created specifically app.py so this is the project main file i'm just opening the copying the project location just to open the anaconda navigator use cd space project location and you enter i'm running python space app.py run the main code just give enter that's it let's wait for the project to complete this project uses the tensorflow platform as well as torch vision platform so you have to enable the torch vision as well as tensorflow to run this project four times the comment box will appear four times you have to give enter that's it third time fourth time so just copy the project location once you copy the project address means Go to the browser just paste it in the browser.
I am just pasting in the project browser who called with the project paste the address here just that's it. So this is the project homepage deep fake face detection using artificial intelligence. We have given the abstract for this project in the bottom of the session.
We are going to use LSTM network here. Just go to the homepage. Just give login.
In the login page you have to use admin admin as a username and password. Admin admin. I'm using this. Just give login. Just login success.
Once login got success means it will go to the next page. So tick fake face detection. Give get started.
So it will go for the detection page. So wait for the detection page to appear. So this is the detection page. Here you can upload any sort of video. You can check whether the video belongs to Deepfakeface creator, VA created video or normal video.
Just I'm evaluating some videos for you. I'm sure paste it in the static folder in the video session. So these are the examples video I will consider okay.
For example I do open one video for you. This is the video for example I don't know whether the video obtained by fake face or not. We don't know. See the video. I'm just uploading the same video here.
Go to the website, just select the same video. Project Navigation, in the navigation part, go to the project folder. I am selecting the same video here.
So this is the video I show, just click the video. As you may see, this is the video which we uploaded. Now we are waiting for the video to upload. We have to waiting for the result to compile. So result already obtained.
it is shown on the website platform so we have to wait for the time okay now the website was in reload mode so now the result update so this is the video you uploaded it's extracted to the frames each and every frames you may see that you may see the green color mark so the green color mark referred to the real face so the result was prediction real overall confidence for this video belongs to real was 91 percent okay so 91 percent this video completely belongs to real you can see the overall frames as well as extractive faces extractive faces also you can see so this is the extracted faces now i am going to upload one more video for you so i am just showing the video uh let's take any video so this video i had to take Express how you feeling it in the student express how you feeling it in the student could make the teacher a better So you have seen the video let's upload this video on our project Just to choose and just upload the same video. This is the video I show just upload the same video here so you may see the comment box the video got uploaded already so let's wait for the video to result to appear so it's was in processing so we have to wait for some again to result to appear so it's processing the frames pre-processing part then extraction part so you may see the red color mark the video completely uploaded by created by aa so this is the fake face but we don't know whether the video belongs to fake or not okay see the things this is the extracted face this is the overall video the complete video belongs to fake so the result was completely fake video okay the overall confidence this 99.9 percent this video completely belongs to fake so do one thing can i show upload the original video Let's check. This is the video which we uploaded. But this is the another video.
Let's see this video. So this is the one more video. This is also the one more video. You may not change the difference.
Okay. You don't know which one is original and which one is fake. So I am just uploading this one.
This ad-hoc-hd. This file I am going to upload. Just go to the project folder. Just upload the same video. this is the another video example just upload this one let's see whether it's directing fake or not okay we don't know it's a original video i hope okay let's see the result so result appeared so this face completely belongs to real so the same nearly exact another video i uploaded for this uh same shirt and all but is shown as a fake video okay so this is the original face this is the original video of that woman okay the result also completely original the confidence percentage of this video belongs to original was 99.79 percent okay this is the top overall result Now let's upload one more video for your safety, okay?
I will upload one more video. Let's take any video. Can I take any video?
Just a minute. Hold on once again. Let's see this video. Let's upload this video.
Yes, perfect. Now, big smile with teeth. Awesome now, Seth. Let's see an upload.
What's happening? Just uploaded this video. you may see it's uploaded here okay we are just waiting for the result to come hit just to wait so it's now extracting the frames and the result and that okay so result came you can see the video completely belongs to a created fake video see that extracted face These are the fake faces.
Overall prediction result was fake. Confidence percentage 99%. Confidently, this video completely belongs to fake.
Let's upload one more video. For example, I have to check one more video. Let's take this one. We don't know whether this is AI created or original or not. Just upload this one and check.
Yeah, I'm not really sure. so i'm just uploading this one just give upload so results same extraction thing going on you have to wait for the results see that so this is the real face this way it's completely belongs to real one okay so like this you can check with any video on there but if i am uploading the rashmega video means what google what youtube will do means it will automatically ban my youtube channel as well as automatically ban this video why because that video comes under the pornography okay so you can upload any video okay once you downloaded this project update this project me you can check with any video or any extra xyz video somewhat anything you can check but perfectly now i'm going to check the overall accuracy of the project to check the accuracy i'm going to the home page and just clicking the accuracy so the overall accuracy of the project we can obtain here okay so the overall accuracy you can see the overall accuracy around 98 percentage for this project let's see the accuracy the overall accuracy of this project was nearly 98 percentage okay as soon as testing accuracy or validation accuracy was 98.2 percentage this is the overall accuracy of this project let's see the overall architecture of the project so this is the overall architecture of the project as we mentioned earlier we are using lsdm here to detect shine as well as class figures okay uh let's go on to the our services as well as more details on our next page okay to get this project please contact iwxpert.com we do provide this project completely Best to press, okay?