Detecting fake images with Machine Learning

With technology accessible at really cheap prices to everyone, it has become easy to edit and tamper with pictures. In this technological era a huge number of people have become victims of image forgery. People these days use technology to manipulate images and use it as evidences to mislead the court. Some are even paid to do this.

Spreading news and images have become so easy these days that a piece of news or an image can spread to millions of people even if it isn’t authentic. Social media is a great platform to socialize, share and spread knowledge but if caution is not exercised, it can mislead people and even cause havoc due to unintentional false propaganda.

While manipulation of most of the photo-shopped images is clearly evident due to pixelization & sloppy jobs by novices, some of them pull it off and make it appear real. Especially in the political arena, manipulated images can make or break a politician’s credibility. Dealing with fake news has become one of the most pressing needs of the digital age.

There are so many fake videos and images flying around on social media sites, it has become extremely difficult to stem the tide. Experts around the world are getting increasingly worried about new AI tools that make it easier than ever to edit images and videos.

The threats:

The most widely used tool for editing images is Adobe Photoshop. Miscreants misuse Photoshop to edit images which could potentially be used to defame people, spread fake news, or to unduly boost the popularity of a politician during election campaigns, thus making it unfair to the others. Photoshop isn’t the only threat to the authenticity of a photograph.

There are plenty of other apps and software available which modify images and make them appear real. A smartphone app called FaceApp, released recently by a company based in Russia, can automatically modify someone’s face to add a smile, add or subtract years, or swap genders. The app also applies “beautifying” effects that include smoothing out wrinkles and, more lightening the skin. Manipulating raw data is not just confined to images, it extends to videos too.

Researchers at Stanford University demonstrated a face-swapping program called Face2Face. This system can manipulate video footage so that a person’s facial expressions match those of someone being tracked using a depth-sensing camera. The result is often eerily realistic. The ability to manipulate voices and faces so realistically could raise a number of issues, such as manipulation of evidence to be provided in a court of law, or using it to defame a politician or celebrity.

The same networks that are created for fun and recreational purposes can be made to generate their own data based on what were able to internalize about the data set they were trained on. It is possible to train such a network to generate images from scratch that look almost like the real thing. In the future, using the same techniques, it may become a lot easier to manipulate video, too.

Given the technologies that are now emerging, it may become increasingly important to be able to detect fake images, video and audio.

The solution:

For now, there hasn’t been much done in the area of detecting fake audio and videos. However, there are many research projects being undertaken to see what can be done about the astonishing spread of fake images throughout the internet. Adobe is aware of the way Photoshop is being misused, and is trying to form an antidote of sorts.

Researchers at the organization have built a model can be used to differentiate between authentic and tampered images using image manipulation detection. The team focused on the three most common tampering methods:

  • Splicing: a form of editing where parts of two separate images are combined
  • Copy-move: a form of editing where parts or objects in an image are copied from one place to another
  • Removal: a form of editing where parts or objects in an image are completely removed, so as to make it look like they were never there in the first place

In order to train the R-CNN (convolutional neural network) to recognize altered and edited images, thousands of images were used as examples. Two different techniques were meshed together to make this neural network. The first technique makes use of an RGB stream, while the second uses a noise stream filter. The below collection of images shows how the final model works:

This is a method that analyses the nature of the image more than it examines the content in it. It is not the perfect answer to the problem of spreading fake images, but it is still a stride towards correctness.

The takeaway:

The very fact the concerns are being raised over potentially altered images, audio and videos draws attention to the fact that people are becoming aware of the risks associated with the immense power of the internet (especially social media) to spread things. As mentioned before, news and images spread before someone even has the chance to examine their authenticity.

There have been reports of manipulated images and news being spread with intentions of defaming politicians, celebrities and public figures. This is utterly unacceptable. This is why it is the need of the hour to find solutions to the rising problem of content manipulation online. Like Adobe has implemented machine learning to create a model that could tell fake images from real ones, there are many others who are dedicated to this cause as well.

There will be other such software developing soon enough, and after that they should be made available to everyone for use, so that they can determine what is true and what is not.