Adobe and UC Berkeley researchers train artificial intelligence (AI) to detect facial operations in images edited with Photoshop software.
While deep visual content is becoming more common and deceptive, the decision is also designed to enable everyone to understand image forensics.
The company wrote in a blog post on Friday: “This new study is part of Adobe’s broader efforts to better detect image, video, audio and document operations.”
As part of the program, the team trained a Convolutional Neural Network (CNN) to discover image changes made with Photoshop’s “Face Away Liquify” feature, which is designed to change facial features such as eyes and mouth.
In the test, it was found that although the human eye was able to judge the changing face within 53% of the time, the trained neural network tool achieved a result of up to 99%.
The tool also identifies specific areas and methods of facial warpage.
A few days after Facebook CEO Mark Zuckerberg and US Speaker Nancy Pelosi’s tampering video were aired on social media and news channels, Adobe performed the execution of detecting facial operations.
“This is an important step in detecting certain types of image editing, and the undo function works very well. In addition to this technology, the best defence will be a seasoned public who knows that the content can be manipulated, usually To make them happy, but sometimes they mislead them,” said Gavin Miller, head of Adobe research.
Adobe Photoshop software was originally released in 1990.