Academia.eduAcademia.edu

CONTENT BASED VIDEO QUERYING

2010, JURNA DASI

CONTENT BASED VIDEO QUERYING Research Proposal Introduction Today, people have access to more video than anytime in the past. Video sources can come from television broadcast, internet video database like www.youtube.com, movie theaters or rental and private video collection. In the past, most of video contain entertainment, but at this time many other type of video like educational video, news video, medical video. As the result of increasing in video number, difficulty in searching desired information becomes a major problem. Video standard representation has improved as demand of video storage. MPEG 7 is a standard format for video that save not only the video frame but also the meta data of the video. The standard has been released on 2002 by Moving Picture Expert Group as ISO Standard no. 15938. MPEG 7 has adopting the XML Schema as the basis of MPEG7 DDL (data definition language) and resulting compliant instances eases interoperability by using common, generic and powerful representation format. . Video is consist of a serialized image called frame. The motion of the video is resulted by replace an old frame with the new one. A smooth moving picture (video) shows 25 – 30 frame per second (25 fps). Video resulted from camera recording, sometime only a single camera but in an advanced video recording they utilize many camera from many angle and background. A key frame is an important frame of a video. A key frame provide an abridged representation of the original video sequence, serving multitude of applications depending the user need. Key frame can store in MPEG – 7 format in XML document. In this paper I will use the collection of key frame collection of images that will use as a data source in image searching. Research Objective This research aims to provide a query to the database system video. This research aims to create a video database model that has a search feature based on the content (content). Content that can be used as a search keyword is the content of image (picture) as well as the contents of the conversation transcript (text). Research Benefits This research is expected to be useful in the field of the future of television broadcasting (TV on demand), the security field to search offenders on CCTV video data and educational videos for learning to search a particular topic or a particular speaker, the field of medical and other fields of using video as one source of information. Hypothesis Preprocessing of the video summary that includes key frames, background, actor transcript and summary transcript can accelerate the search process and increase the accuracy of the video query response video. Methodology This research goal is to provide the system to manage video database that fast to search using both textual and visual features. In this research I divide a video into two features, voice feature and visual feature. Voice feature will convert into textual data. Visual feature can be extracted from a series of frame. Each frame is a still image. Visual feature will be extracted into several important frame, it generally called as key frame. Other extracted result is actor, especially for video that contains some human, it will extracted as actor. There are several processes in this proposal in order to form multimedia database model. First process is try to classify video from visual features. This process will identify key frame. Key frame will extracted into actor and background. After actors and background extracted, they will be stored as attributes of a video. Second main process is extract voice from video. The voice will filtered to remove noise from voice data. Clean conversation voice will be modified into text. In order to avoid big size of text, it will processed to several important word (key word). The text and keyword will store as video attributes. MPEG 7 format can store those information as XML tags. Video database with several additional attributes from visual feature and voice features will be the basis for the retrieval process. Figure 1 Block Diagram Content Based Video Querying Video retrieval process will utilize video database model. There are four kind of input that user can use as keyword to the system. User can use text keyword, actor image, background image as query to the system. A user can use only one input or combination of several input. References 1. Jose M. Martinez, Rob Koenan, Fernando Peirera. MPEG-7 the Generic Multimedia Content Description Standard, Siemen Corporate Research 2002. 2. Herman Furntratt, Helmuth Neuchmidd, Werner Balier, MPEG 7 Library, Joanneum Research, 2009 3. M Khansari, H.R Rabiee, M. Assadi, M. Ghanbari, Object Tracking in Crowded Video Scenes Base on Undecimated Wavelet Feature and Texture Analisys, EURASIP Journal of Advanced Signal Processing Volume 2008, 2008 4. Darin Brezeale, Diane J. Cook, Learning Video Preferences Using Visual feature and closed caption, IEEE Multimedia, July - September 2009