Taking into consideration how sophisticated technology became, creating modified or altered images and videos has never been easier. It became common to see this type of media involving celebrities and politicians.
However, the recent idea created might give an edge to people trying to copyright their material. The test prototype showed an increased chance of detecting altered videos and images from 45 percent to over 90. The best thing is that the process won’t sacrifice the quality of the image or video.
A research assistant professor, Paweł Korus, teaches at the New York University in the computer science and engineering department. His idea involves replacing the standard pipeline development with a new neural network, which is a form of AI. This way, carefully crafted “artifacts” will be implemented in the original image at the moment of acquisition. Furthermore, these artifacts are especially sensitive to any form of manipulation.
Before this, the images used standard watermarks that were obviously flawed. The new AI will detect any kind of manipulation and even their character. The idea behind it is to implement this whole process for in-camera embedding. Also, the image will be able to survive any form of distortion online photo sharing services might apply.
If they managed to perfect this, the camera that produces the image or video would be able to create an image that is more sensitive to any kind of tampering, and all modifications will be detected easily. Furthermore, these artifacts can even survive post-processing, but they are somewhat fragile when it comes to image modification. Basically, if you alter the image, the watermark or artifact will break.
Almost every other form of detecting fake content is complicated and won’t give such quality results. This new technology will use machine learning that already takes place with every photo created by a smartphone. For example, all the adjustments for stabilizing or low-light or are created by AI and machine learning.
That will give room to new technology that can gradually replace the standard imaging pipelines. Nasir Memon, who is a co-author of the idea with Korus, said that this could change the capabilities of devices in the future (or next-generation) when it comes to both the authenticity and integrity of media. The whole program is currently in testing, and the solution is open-source. You can access it for more details, and the paper will be presented in Long Beach, California, at the Conference regarding Computer Vision and Pattern Recognition.