TY - GEN
T1 - Advertisement Extraction Using Deep Learning
AU - Madi, Boraq
AU - Alaasam, Reem
AU - Droby, Ahmad
AU - El-Sana, Jihad
N1 - Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021/1/1
Y1 - 2021/1/1
N2 - 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.
AB - 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.
KW - Ads extraction
KW - Loss function
KW - Regression model
KW - Segmentation model
UR - http://www.scopus.com/inward/record.url?scp=85115346008&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-86159-9_6
DO - 10.1007/978-3-030-86159-9_6
M3 - Conference contribution
AN - SCOPUS:85115346008
SN - 9783030861582
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 81
EP - 97
BT - Document Analysis and Recognition – ICDAR 2021 Workshops - Proceedings
A2 - Barney Smith, Elisa H.
A2 - Pal, Umapada
PB - Springer Science and Business Media Deutschland GmbH
T2 - International Workshops co-located with the 16th International Conference on Document Analysis and Recognition, ICDAR 2021
Y2 - 5 September 2021 through 10 September 2021
ER -