Advertisement Extraction Using Deep Learning

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.