Semantic video retrieval using deep learning

Abstract

Contemporary video retrieval systems are ineffective when it comes to retrieving a video based on the content inside it. The systems rely on accurate tags and titles of videos for their accurate retrieval. However, mischievous personnel can easily register false tags or titles of the videos, significantly reducing the capabilities of a video retrieval system to accurately retrieve videos. This paves way for spam, inaccurate content retrieval and difficulty in analyzing the hours of content that is generated each minute. The system being proposed in this project will allow users to retrieve videos through the actual content inside that specific video. Using this system, uses will remain protected from being spammed by people who put in wrong tags or titles to attract users. The user base of this system is not just limited to home users who watch videos. Monitoring agencies, search engines and other institutions can use this system to create monitor content, create a more reliable database of videos and perform other effective tasks, which were previously impossible.