International Research Journal of Engineering and Technology (IRJET)
Volume: 04 Issue: 07 | July -2017
www.irjet.net
e-ISSN: 2395 -0056 p-ISSN: 2395-0072
Video Content Identification Using Video Signature: Survey Tejashri Shinkar1, D. B. Hanchate2 1 Student
of ME-II, Department of Computer Engineering, VPCOE, Baramati, Savitribai Phule Pune University, Maharashtra, India 2Department of Computer Engineering, VPCOE, Baramati, Savitribai Phule Pune University Maharashtra, India ---------------------------------------------------------------------***--------------------------------------------------------------------Abstract – The amounts of videos generated by people are increasing day by day. At each minute billions of videos are uploaded over the network. The videos are different types duplicate, edited, pirated etc. There are very few tools are available to find out near duplicate videos. Existing methods are not finding out video segment from larger unrelated video. These methods generate a signature of three types like spatial, temporal and spatio-temporal. This paper proposed a method which extract frame level features through this create a temporal signature. Using this signature it accurately detects a video segment which is embedded into larger unrelated video.
High precision and high speed video identification is achieved through the video signature without embedding ID information in the content. The Video Signature Tools, which standardizes an interoperable descriptor for video identification. This system performs in three steps: 1) Video Signature Extraction 2) Video Signature Compression and 3) Video signature matching. 2. PREVIOUS WORK Literature survey includes different design issues for creating video fingerprint/video signature. It is also focus on different strategies for localization and identification of a video embedded in unrelated video content. The Video signature can be used for near duplicate clip detection. Basically the video signatures are based on the frame level features such as keypoint based, block based or global. The problem of identification of video content embedded in longer video content is studied by Hampapur et. al. [2]. In this paper, ordinal signature which is based on blocks achieves highest performance. To calculate signature first frame is divided into blocks. The comparative study between different types of signature done by Hampapur et. al[2]. In his work he considered three types of signatures as follows:
Key Words: Content identification, Video Signature, Content localization, Video Identification, CBCD technology, BCS, LBP, Spatio-temporal features 1. INTRODUCTION In the mass media industries; the data volume is growing enormously. There is also advancement in the telecommunication and in the internet data transfer. Also infrastructure of multi-platform content delivery is advanced which accelerate the transmitting speed and increased volumes. Due to these conditions, video contents are generally managed using metadata like keywords, thumbnail images, preview videos, etc. Generally this provides significant volume of metadata search results through the increase in the data volume and the circulation of video contents. Due to this it is difficult to identify video content or to search a specific scene manually by visual inspection. To solve this problem, paper proposed a technology that identifies video content automatically and efficiently by managing it as metadata. The proposed system focuses on the first to design tools for video fingerprinting that will provide high robustness over common editing operations like cropping, labelling, morphing etc. Also detect the particular video content that will be embedded in longer unrelated video. Video identification technology analyses each frame of the video content and extracts a descriptor which is unique. This unique descriptor is called video signature which is used to identify identical video scenes.
Š 2017, IRJET
|
Impact Factor value: 5.181
1. Ordinal Signature: For signature calculation first divide each frame into blocks. Then calculate mean intensity for each block and sort them into ascending format. Rank vector will be calculated that is used as the feature of the frame. The procedure will be as mentioned in the paper [8][9]. Best performance is reported for shorter queries. This paper did not report the matching segment localization accuracy. 2. Motion Signature: For each block in frame , select image patch at block center. Find the SAPD (Sum of absolute Pixel Differences) at each point in search neighborhood frame. To detect match location minimum SAPD is considered. For best
|
ISO 9001:2008 Certified Journal
|
Page 746