Advertisement Extraction Using Deep Learning

Boraq Madi, Reem Alaasam, Ahmad Droby, Jihad El-Sana

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


This paper presents a novel deep learning model for extracting advertisements in images, PTPNet, and multiple loss functions that capture the extracted object’s shape. The PTPNet model extracts features using Convolutional Neural Network (CNN), feeds them to a regression model to predict polygon vertices, which are passed to a rendering model to generate a mask corresponding to the predicted polygon. The loss function takes into account the predicted vertices and the generated mask. In addition, this paper presents a new dataset, AD dataset, that includes annotated advertisement images, which could be used for training and testing deep learning models. In our current implementation, we focus on quadrilateral advertisements. We conducted an extensive experimental study to evaluate the performance of common deep learning models in extracting advertisement from images and compare their performance with our proposed model. We show that our model manages to extract advertisements at high accuracy and outperforms other deep learning models.

Original languageAmerican English
Title of host publicationDocument Analysis and Recognition – ICDAR 2021 Workshops - Proceedings
EditorsElisa H. Barney Smith, Umapada Pal
PublisherSpringer Science and Business Media Deutschland GmbH
Number of pages17
ISBN (Print)9783030861582
StatePublished - 1 Jan 2021
EventInternational Workshops co-located with the 16th International Conference on Document Analysis and Recognition, ICDAR 2021 - Lausanne, Switzerland
Duration: 5 Sep 202110 Sep 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12917 LNCS


ConferenceInternational Workshops co-located with the 16th International Conference on Document Analysis and Recognition, ICDAR 2021


  • Ads extraction
  • Loss function
  • Regression model
  • Segmentation model

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'Advertisement Extraction Using Deep Learning'. Together they form a unique fingerprint.

Cite this