Image editors have been using tools such as Adobe Photoshop to edit images for decades, sometimes to the extent of complete fabrication.
This is a concern, especially with the presence of social media platforms, which allow users to share shocking and viral content without verifying the facts.
Adobe's AI Can Detect Fake Images
Adobe revealed that it is developing a new software that uses artificial intelligence to spot manipulated images.
Whenever an image is altered or edited, editors always leave behind some form of evidence that reveals that the image is modified. Adobe's research team, including senior researcher Vlad Morariu, recently published a detailed analysis on how it is training the AI system to recognize doctored photos using this evidence.
How Is The AI Trained?
The team pointed out that it focuses on three main types of image edits that are often used to create fake images. These "tampering techniques" include splicing, copy move, and removal. Splicing is when an object is added from one image to another, copy move is cloning objects within the same image, and removal, as the name suggests, is the removal of an object from an image.
Researchers at Adobe teamed up with researchers at the University of Maryland and fed the AI system thousands of fake images in order to teach the software what exactly to looks for. They trained the software to detect the modification in an image and flag the area that has been altered, and the AI can achieve this within seconds, according to Adobe.
What Techniques Does The AI Use?
One method that the software uses to spot phony images is by looking for discrepancies in the red, green, and blue color values of pixels in altered images.
The other method that the system uses to find fakes is by generating a noise map of the picture. Noise, the random variations of color and brightness caused by a camera's sensor, are unique to cameras or photos. Because of its unique pattern, objects that have been spliced in from another image will stand out when a noise map is generated, allowing the AI to pick up on the inconsistencies.
Adobe noted that although these techniques are far from perfect, they "provide more possibility and more options for managing the impact of digital manipulation, and they potentially answer questions of authenticity more effectively."