Logo Detection Plagiarism Using Generative Adversarial Network

Logo, as a graphic form and visual representation of a company, organization, or brand, has been widely utilized and disseminated across the internet. This accessibility makes logo easily replicable, allowing for the imitation and reuse of concepts or forms for creating new logo. Therefore, it is required to have a system that is able to analyze and identify the similarity between logos. In this study, logo plagiarism detection is proposed by using Generative Adversarial Network based on CycleGAN approach. The logo training process involves generator and discriminator networks as part of the GAN process to recognize the logos characteristics. The trained GAN model is tested by generating logo images using the model weights and inputting comparison logos to test their similarity through Template Matching method for plagiarism detection. From the generated test logo results, the logo produced by the generator closely resemble the original logo. The training results using GAN based on CycleGAN approach yielded average loss value of Generator A 6.3572, Generator B 6.1541, Discriminator A 0.6378 and Discriminator B 0.7186. The lowest logo plagiarism analysis value was obtained with an accuracy of 20.11% for logos with gradual changes in shape and different perspectives. The highest accuracy of 93.40% was achieved for logos with no changes at all.