Kaptur

How to catch a fake

An image is not fake until it is identified as such. To do so, it has to be compared to a ground truth and fail.

The ground truth can be the event itself or, for lack of a time machine, the original photograph of the event recorded at the time.

There are three types of fake photos: 

Real image ( after the Japanese earthquake), wrong context

 

Example of an altered image. Graffiti and broken head are alterations.
Portrait generated by a GAN (generative adversarial network). This person is not real.

Each is an attempt to convince viewers of the reality of an event by means of deception.

 In other terms, fake photos falsify an event in 3 ways:

  1. The event did happen, but not when and how is it is described—the out-of-context photo.
  2. The event did happen, but not how it is visible in the image—the altered-content photo.
  3. The event never happened—the 100% created image.

So bar the viewer being an actual witness of the event and having a perfect memory of it, which is a very rare, if not impossible occurrence, the best way to proving an image is fake is comparing it to the original photograph.

Both 1. and 2. use an original image as the base to their deception. For 3. things are more complicated as we will see later. If the deceiving picture can be compared to its original unaltered version, including associated metadata, it can be debunked. Thus establishing the original image as ground truth is primordial to exposing fake photos.

Popular methods for 1. and 2. are:

Each method has advantage and shortcomings:

Blockchain carries an aura of impenetrable and hack-free security. However, it has its flaws.

First, the registered image is not the image but a hash of the picture. Those vary depending on who creates them. The same image can have a different hash. Thus while the blockchain might be an independent network of servers, the creation of the input information is proprietary to those who create it. Not independent, not free, and certainly not impervious to influences.

Also, a hash is not an image. The link between the hash and its corresponding image and original caption is also held in a private proprietary server.

In other words, blockchain is not as secure and certified as it claims. Example

Keeping a core catalog. The most popular solution today. Risks and limitations are obvious. Servers are private, can fail, and their content can be altered or deleted. It can also be hard to retrieve corresponding files because of poor indexing and search capabilities. However, the original image and associated original metadata are associated in one place, establishing a ground truth. Example

Imatag’s invisible watermark can spot an image alteration without having to compare it with the original file.

Invisible watermarking. New and very promising. Certification travels with the image. If an image element is altered, it is easily detectable ( part of the watermark is missing) An instant number 2 killer. To debunk out of context images, the unique watermark can easily trace back to the original file and its original metadata. Ground truth can quickly be established if linked to a core catalog. Example

Date certification. Reverse visual search is the poor people tool to identify fake photos and one of the most in use. Dropping it into a Google image search will reveal ( almost) every instance of its publication. Finding earlier, if not the earliest publication can establish the ground truth photo. Issues here are various:

Ironically, photography has evolved mostly as a tool to alter reality rather than to reproduce it with fidelity. Capturing it with the highest accuracy possible only to alter it with the most advanced computerized imaging options. In its quest for popularisation, it has used modification as its most potent seductive attribute. Today, image falsification in all its forms is a multi-billion dollar industry. Not so much for reproduction fidelity.

There is much catching up to do.

Author: Paul Melcher

Paul Melcher is the founder of Kaptur and Managing Director of Melcher System, a consultancy for visual technology firms. He is an entrepreneur, advisor, and consultant with a rich background in visual tech, content licensing, business strategy, and technology with more than 20 years experience in developing world-renowned photo-based companies with already two successful exits.