The creators of Photoshop are investigating how machine learning can help detect manipulations of images to your photo retouching program helped to popularize, to the point that in the common speech to retouch a photo, many times it is called photoshopear. By so doing they have presented an article of investigation to the days of machine vision, CVPR, held last week in Salt Lake City.
The technique known as machine learning consists in feeding a neural network with huge data sets to train her, so that when you encounter another element that shares characteristics with the original data set can identify you correctly. In this case, the data sets have been pictures that they know they are retouched in three concrete ways: when an object is added to another photo, when deleted, and when you copy and paste a part of the same photo, just like what happened in the photograph manipulated by Iran that leads the article. That is to say, that the classic use of Photoshop to smooth the skin of the models or improve its measures would not be included.
According to The Verge, a spokesperson for Adobe has said that this is a research project that is in its early stages, but that in the future the company wants to play a role in “developing technology to help locate and verify the authenticity of digital images”. It remains to be seen if this type of developments are now more critical for Adobe in the future, because at the same time also has projects in artificial intelligence that create a background sound for realistic videos without sound, fixed bugs in the image generating even objects that are not in the original or remove objects that we don’t want in our pictures. It seems that, regardless of what you say, Adobe invests more in tools to retouch in tools to detect the retouching.