In this vertical we unleash the power of advanced vision technologies to detect and counteract falsified content for enhanced trust and authenticity in digital media.
a. Fake Visual Content Detection
As adversarial learning progresses, generating realistic fake visual content
becomes more widespread, prompting the development of detection techniques.
In this project we explore two-stream convolutional neural networks, leveraging
both frequency and spatial domain features for improved fake detection.
Related Publications
b Learning to Localize Image Forgery
This study aims to develop algorithms for authenticating visual content and
introduces novel approaches using channel attention convolutional blocks within
an end-to-end learning framework. By extracting attention-aware multi-resolution
features in both the spatial and frequency domains, the algorithm effectively
localizes forged regions in images.
Related Publications