Adobe sees fake images as such a big problem that, together with researchers from the University of California, a tool has been developed that will make it possible to recognize photos of faces that have been manipulated using Adobe&s Photoshpos Face Aware Liquify function, a kind of subtle distortion of faces.
The new method, the otherwise digital forensics specialist&s only ability to detect manipulations of digital images, is to be made available to a broad public in the form of a tool to curb the abuse of subtle changes to faces.
The new tool is only part of Adobe&s research efforts to detect manipulation of videos, images, or audio recordings. Last year, a method was introduced to detect three common image manipulation techniques, namely splicing (the combination of parts of two different images), copy move (in which objects in a photograph are moved or cloned from one location to another), and removal (in which an object is removed from a photograph and the resulting space is filled).
Manipulated photo, discovered manipulations, proposed undo and the original photo
Each specific method of image manipulation leaves behind typical traces that can be detected by precise analysis. By training a neural network with thousands of examples of such manipulations, it can also be used to detect manipulations of new images.
The same was done in the recent case, only that the training material here consisted of pairs of original images and faces (and which were manipulated by humans) changed by Face Aware Liquify. The algorithm was then able to correctly detect 99% of photos of changed faces as manipulated. It can even show the regions that have been manipulated and try to restore the original state.
The research is continued in order to detect more complex manipulations, e.g. on the body or the skin. Adobe sees itself as obliged to provide not only tools which enable the manipulation of digital images, but also those which show such manipulations afterwards, in order to enable a certain transparency.