Blurring of Brands with Faster R-CNN


Creative Commons License

Güler D., Karslıgil Yavuz M. E.

II. INTERNATIONAL CONFERENCE ON INNOVATIVE ENGINEERING APPLICATIONS, Muş, Turkey, 20 - 22 May 2021, pp.591-598

  • Publication Type: Conference Paper / Full Text
  • City: Muş
  • Country: Turkey
  • Page Numbers: pp.591-598
  • Yıldız Technical University Affiliated: Yes

Abstract

This paper focuses on techniques of blurring advertising objects in video with deep learning. The project designed and implemented a system that blurs the advertising content of the video with deep learning methods. Faster R-CNN was used as a deep learning method for detecting and recognizing objects. Many problems were encountered in this study. First, the 9 different classes were collected to train the system. After preparing the dataset, Faster RCNN was used to train the system with layered architecture. The Faster R-CNN method is very effective compared to the other methods R-CNN and Fast R-CNN methods. The dataset is divided into %20 test and %80 train. After the training, the success rates of the pictures in the test folder were observed. Different success rates have emerged for each class. Finally, the object containing the advertising product was blurred after the operations. Gaussian filter was used for blurring. The colors of the object found with the Gaussian filter are softened. This produces a blurred image on the screen. The success rates were measured with the obtained images