Video-Sense: Content-based Video Copy Detection System

Abstract

Recently, multimedia sharing has become an integral part of the entertainment and the videos can be shared by users on the internet through social networking websites such as YouTube, Facebook and Twitter. The sudden increase in the volume of multimedia also causes problem of copyright infringement. Therefore, copyright management has become an important issue in an enterprise. The main objective of this project is to develop a video copy detection system which is based on content fingerprinting. The system is called Video-Sense. Temporally Informative Representative Images (TIRI) has recently been introduced as an alternative to watermarking. Its aim is to identify a video based on its content. The efficiency of video fingerprinting algorithm including TIRI-Discrete Cosine Transform (DCT), TIRI-Perceptual Hash (PHASH) and TIRI-Average Hash (AHASH) [4] will be evaluated in this project. These algorithms are using a frame to represent a video segment which contain temporal and spatial information about the video. Novel fingerprint Fast Video Match Strategy, which invented by Ph.D student Liu Mengyang, will be used to search and match the query with video database. The project will evaluate the differences between new Fast Video Match Strategy and sub-fingerprint based inverted file searching algorithm. The implementation of the Video-Sense is based on the python code of a Master student project. The Video-Sense system provides two main platforms for the users including web-based and Windows application. For web-based, video fingerprint need to be extracted and store them in MySQL and hash table file on server side. The users can upload a video clip to server for searching the video database. For Windows application, users can extract the fingerprint and store them in video database by providing a location path of the videos. Users can provide a query video path for searching the video database and matching results will be returned to them.