The rapid advancements of image editing technologies and their increasing accessibility makes it easy to create and distribute very realistic and convincing fake images. While this benefits many fields such as movie productions, advertisements, and virtual reality, this very same technology can also be used for malicious intentions such as manipulated media pornography, identity theft, fake news, and online harassment. In this talk, I will present some promising works on detecting fake images made from splicing or compositing images. Typically, these forgeries are cleverly hidden within the semantic contents of the image, making it difficult to detect through visual inspection. Therefore, in contrast to standard image classifiers that look at high level visual features, we instead show how to extract low-level imperceptible noise and use them as an indicator for forgery.
Daniel Stanley Tan is currently a Postdoctoral Researcher under Professor Jun-Cheng Chen at Academia Sinica. His research mainly focuses on computer vision, generative models, and more recently image forensics. He obtained his Ph.D. from National Taiwan University of Science and Technology under the supervision of Professor Kai-Lung Hua. Prior to this, he was an AI Research Engineer at Inventec Corporation where he worked on defect detection for industrial products. He also held a faculty position at De La Salle University, Philippines, where he taught computer science related courses.