Fake Investigation Department

Our Sherlock Holmes against Fakeness

Comment

Inspiration

Nowadays people rely mostly on news articles, social media, and many other online platforms to get knowledge about their surroundings. Some people can easily take advantage of these online platforms to spread fake news and images about sensitive topics which can also cause riots and stress in society. So we decided to make a platform to deal with this problem.

What it does

Our product has two services:-

  • News curator:- It will help users to identify the neutrality and authenticity of news articles related to the topic provided to the curator along with the original source of the news article.

  • Image Investigator:- It will help users to find authenticity in the image provided to Investigator by providing them with a binary mask image highlighting manipulated regions in the image.

How we built it

For ML models Image Investigator

  • We read research papers on Image manipulation and learned some new techniques for image manipulation detection and localization
  • We also searched for different datasets related to image tampering and had to increase it by combining different datasets
  • We combined different techniques from different papers to built our algorithm

News Curator

  • We compared different algorithms like SVC, Random Forest, Naive Bayes etc.
  • We decided to use Max Voting Algorithm which automatically decides the best model to use by comparing accuracy

Later we created seperate frontend and backend for each of the services using React (frontend) and Flask (backend) to create our combined product - Fake Investigaton Department

Challenges we ran into

We have faced quite many bugs in this project. We overcome these hurdles through research and learn new technologies. Some are worth mentioning bugs and the hurdles are as follows.

  • To complete the dataset for Image Tampering. We have to merge different datasets for specific tasks like copy-move, splicing, and removal, etc, pre-process them accordingly and create a complete dataset that can be used to train our model for fake image analysis successfully.

  • As our backend and frontend were created on different machines. We used Ngrok to save time and host both of them together.

  • On the frontend side, we have used React.js that increases the number of files. To handle multiple files easily we used Webpack. Webpack is a tool that lets you compile JavaScript modules, also known as module bundler. Given a large number of files, it generates a single file (or a few files) that runs your app.

  • For running our deep learning model, we faced the problem with different versions of Tensorflow and Keras, and using various Stackoverflow issues we solved those compatibility issues.

Accomplishments that we're proud of

  • Completing the product on time before submission
  • Having good accuracy for both the News Curator and Image Investigator
  • Getting to learn a lot about frontend, backend development as well as deep learning

What we learned

  • Getting better at converting research papers to code
  • Learned a lot about using React and Flask
  • Integrating Backend with ML model

What's next for Fake Investigation Department

  • Adding more services related to dealing with fakeness like Fake Reviews, Fake Videos etc.
  • Increasing the accuracy of deep learning and machine learning models used in the product
  • Monetization of the services by creating a paid subscription for a small fee
  • Adding advertisement to the product (but not too many to create problems while using)

Built With

Share this project:

Updates