Story Based Video Retrieval using Deep Visual and Textual Information
Ahmed M. Hassan, Summra Saleem, Muhammad Zeeshan Khan, Muhammad Usman Ghani Khan · 2019
From the past few decades, popularity of online news videos have been increased day by day because of rapid advancement in technology and internet availability. At most all the news channels have been uploading their news in the form of headlines and bulletins on world wide web. To watch complete news headline and bulletins of a specific TV channel is very time consuming. Users want to watch the news of their interest instead of watching the whole news-headlines and news-bulletins. This paper deals with the retrieval of a specific NEWS category like World, political, weather etc from the news headlines based on user requirements. Our proposed system is based on deep convolution neural network. Accuracy and effectiveness of proposed architecture have been evaluated using 2D convolution neural networks. The dataset has been collected from different news channels and sorted according to the categories. We have collected 200 videos for eight categories of news. Our trained model is able to recognize the categories of a news with 91.7% accuracy. The distinction between news categories with such high accuracy on the locally prepared dataset shows the novelty in research. The dataset is collected from 6 Pakistani news channels. Dataset and live Web application of this methodology will be made publicly available for research community.