Identifying and understanding the Difference between natural Images and computer-Generated images (IDIG)

 

Identifying the difference between natural and computer generated images

It is often very difficult to distinguish between natural and computer-generated (CG) images by human naked eyes, see examples at the Fake or Foto website. Our first objective of this PERSYVAL-Lab exploratory project is to develop new and effective methods to identify computer generated fake images that do not reflect real-world scenes. We plan to derive distinctive features, either hand-crafted or automatically extracted, for the purpose of developing a forensic method that can expose both full-sized CG images and CG patches of small sizes. We are particularly interested in developing a deep-learning-based forensic method to distinguish between natural and CG images.

 

Understanding the difference between natural and CG images

In the second part of this project, we will focus on understanding the main reasons for the statistical difference between natural and computer generated images. More precisely, we will conduct experiments to investigate the influence of each key element of the rendering algorithm on this statistical difference. With the learnt knowledge from the experiments, it would be possible to develop a learning-based post-processing method to improve the realism of a CG image or even a more effective rendering algorithm that can produce visually more pleasing graphical images. Besides the scientific objectives, this exploratory project would also serve as a starting point for establishing solid collaborations between the two participants of the project, located respectively in Grenoble and Beijing.

Participants

Kai Wang, CNRS researcher, GIPSA-lab, Grenoble
Dong-Ming Yan, Research Professor, Chinese Academy of Sciences, Beijing
Weiza Quan, Ph.D. student, Chinese Academy of Sciences, Beijing

Minh Kha Nguyen, Master intern, Ensimag, Grenoble

 

News

Our paper got accepted and published at IEEE TIFS (link to the paper).

Our paper got accepted and published at Elsevier's journal of Digital Investigation (link to the paper).

Weize pays a 6-month visit to GIPSA-lab from the 1st March 2017 and works on the IDIG project.

Minh Kha starts his M1 internship on the IDIG project at GIPSA-lab from the 30th January 2017.

 

 

Return to ''Projects'' page