Irdeto protects digital platforms and applications for media & entertainment, automotive, and IoT connected industries. Irdeto solutions and services enable its customers to protect their brand and revenues, create new offerings and fight cybercrime. Irdeto is a part of Naspers. Founded in 1915, Naspers is a global internet and entertainment group and one of the largest technology investors in the world. Operating in more than 120 countries and markets with long-term growth potential, Naspers builds leading companies that empower people and enrich communities. It runs some of the world’s leading platforms in internet, video entertainment, and media.
The goal of the project is to propose a solution that uses Deep Learning to covertly mark image data and detect the hidden mark.
Task 1.
Create a system consisting of neural networks named Alice and Bob. Alice takes as input image (I1) and n-bits value (V1) and outputs image (IM). Bob takes as input image (IM or I1), outputs n-bit value (V2).
Alice wants to minimize distortion introduced into the image (e.g. PSNR(I1,IM) > threshold). Alice and Bob want to minimize Bob’s detection error (ideally, V2=V1 for IM and V2=0 for I1).
Task 2.
Create a system consisting of neural networks named Alice and Bob. Alice takes as input image (I1) and n-bits value (V1) and outputs image (IM). The image (IM) is than transformed by mod module into IMT . Modification may include cropping, padding, rotation, scaling, blurring, de-colorizing, re-encoding. Bob takes as input image (IMT or I1), outputs n-bit value (V2).
Alice wants to minimize distortion introduced into the image (e.g. PSNR(I1,IM) > threshold). Alice and Bob want to minimize Bob’s detection error (ideally, V2=V1 for IMT and V2=0 for I1).
Task 3.
Create a system consisting of neural networks named Alice, Bob and Eve. Alice takes as input image (I1) and n-bits value (V1) and outputs image (IM). The image (IM) is than transformed by mod module into IMT . Modification may include cropping, padding, rotation, scaling, blurring, de-colorizing, re-encoding. Bob takes as input image (IMT or I1), outputs n-bit value (V2). Eve takes as input image (IM or I1) and outputs a) 0 for I1 and 1 for IM , b) n-bit value (V3).
Alice wants to minimize distortion introduced into the image (e.g. PSNR(I1,IM) > threshold). Alice and Bob want to minimize Bob’s detection error (ideally, V2=V1 for IMT and V2=0 for I1) and to maximize the error of the “optimal Eve”. Eve wants to a) differentiate between IM and I1; and b) minimize detection error (ideally, V3=V1 for IM).