With Adobe Photoshop that came about 28 years ago, analog photos were brought into the digital space. Now, we can edit images to give a new dimension to the artistic expression. Editing images also become useful to trace missing people and even preserve our history. There is also another dimension to editing images. People edit images for deceptive purposes. Here is how you can Spotting Image Manipulation.
Together with creating digital media, Adobe is using artificial intelligence to enhance the authenticity in the digital media.
Vlad Morariu, a senior research scientist at Adobe has taken up the challenge of image manipulation. This is part of the DARPA Media Forensics program he has undertaken.
Spotting Image Manipulation with the help of various tools
According to him, there are already tools in existence that help in spotting image manipulation. There are file formats which contain metadata that store information. This data shows how the image is captured and manipulated.
Forensic tools can help in spotting image manipulation by checking out the noise distribution, lighting, and other pixel values of the image.
With the help of Watermarks, the originality of the image could be ascertained and help in spotting image manipulaton.
All these tools do not provide a conclusive understanding for spotting image manipulation. One cannot get a complete sense of the authenticity of the image and neither is it practical for very situation. Some of the tools may require a lengthy execution and analysis while other tools may just not be function.
Finding Image Manipulation with the help of AI
Vlad Morariu proposed the use of technologies like artificial intelligence and machine learning. These could be used easily, they will be reliable and also help in quickly spotting image manipulation. They can detect if any part of the image is manipulated and if it is so, then what aspects had been modified.
Vlad Morariu started his research fourteen years ago and it continued while he was a PhD student in computer science at the University of Maryland.
According to him, manipulation of the images could be due to splicing, copy-move and removal.
In splicing, parts of two different images are combined.
In copy-move, the objects in the image are either moved or cloned from one place to another.
And In the case of removal, an object is removed from the image and filled in.
Every time an image is manipulated, there are certain clues that are left behind. There could be strong contrast edges, deliberately smoothed areas or differing noise patterns. Most of these manipulations may not be visible to the human eye. On closer analysis at the pixel level or by applying filters, these manipulations will be highlighted.
With the help of AI, spotting image manipulation is far more successful. AI identifies the type of manipulation and also the specific area where the image is altered.
The two techniques used are the RGB stream and the noise stream filter.
The RGB stream changes to red, green and blue color values of pixel to help in spotting image manipulation.
Many images and cameras have a unique noise pattern. One can detect noise inconsistencies between the original and manipulated areas. This is mainly in the case where the image is combined from two or more photos. This technique is also used to help in spotting image manipulation.