PIP Detection And Its Application In Video Copy Detection

Content-based video copy detection (CBCD) aims to determine whether the given video stream includes some content derived from some original video clips. CBCD can be applied to lots of applications, such as copyright protection of digital media, semantic relationship mining, clustering of video archives. Generally, instead of directly inserting the original video clips into video stream, the authors pre-process (or transform) those original video clips before using them. As tough video transformation, picture in picture (PIP) is frequently used when making video works. It is vital for video copy detection to quickly and precisely detect and localize the PIPs in video streams. Most of existing methods generally treat the whole video stream as a detection unit. That is, if PIP occurs in a video stream, it will last from the first frame to the last frame. However, multiple PIP cases frequently occur in real-world application scenarios, such as news video, sport video. Therefore, new detection algorithms should be designed to handle those cases. In this thesis, we focus mainly on the detection of multiple PIPs in unbounded video stream. The main contributions in this thesis include:1. Propose a spatio-temporal multiple PiPs localizing algorithm. By discovering a successive-group prior of both overlapped PiP and non-PiP clips based on the statistics of single PiP detection accuracy, we introduce a series of geometrical and temporal constraints into our PiP detection framework so as to handle both the spatial and temporal localizing problems. The experiments demonstrate that our scheme can efficiently localize the spatial positions and precisely detect the starting and ending time stamps of PiP clips.2. Design and implement a video copy detection system. To show the effectiveness of the proposed multiple PIPs detection algorithm, we design and implement a PIP-based video copy detection system, which is based on B/S architecture. Though it is only an experiment system, the design and implementation of system follow the rules of software design.