BrandOff: Content Obfuscation of Copyrighted Branding or Trademarks in Images Using StyleGANv2
Prajwal Jadhav1, Vaibhavi Lambole2, Ajay Pandita3, Govind Rajput4, and Sharmila Wagh5
1, 2, 3, 4 Dept. of Computer Engineering, M.E.S. College of Engineering, India
5 Professor, Dept. of Computer Engineering, M.E.S. College of Engineering, India
Abstract
Logo design is an essential aspect of branding that can be a powerful symbol of a company’s identity. However, the use of logos is heavily regulated and using copyrighted logos without permission can lead to significant legal consequences. This has led to growing interest in developing techniques for generating logos that are both unique and legally permissible. In this study, the authors explore the use of StyleGAN v2, a trailblazing generative model, to generate logos. The authors propose an approach that involves training the StyleGANv2 model on a dataset of existing logos to generate new logos that are visually similar but legally permissible. The authors then use YOLOv5, a cutting-edge object detection model, to detect well-known logos in images before obfuscating them with the generated logos. The authors conducted experiments on a dataset of images containing copyrighted logos and compared the original images to images with obfuscated logos. The results suggest that using generated logos to obfuscate well-known logos can be an effective means of avoiding copyright infringement while preserving the visual consistency of images. This research has important implications for the field of logo design and copyright law and demonstrates the potential of generative AI and object detection for creating unique and legally permissible logos for various applications.
Keywords— generative adversarial network; styleGAN; styleGANv2; yolov5; copyright obfuscation
