Distance education in Content-Based Video Retrieval
【Abstract】 The video is an important content of distance education, due to video their own characteristics, the urgent need for content-based video retrieval research.
This paper first proposed the inherent characteristics of video, followed content-based
on their analysis, and describes the status of research on video at home and abroad.
Key words video; content-based; Search
1, distance education, the need for content-based video study
Digital video in distance education in the information system is a multi-media
teaching important data type is characterized by large volumes of data and informative large. If a 24mm × 36mm (commonly known as the 35mm) color photographs, if the
spacing of 12um scan, then the formation of three color digital image. Each piece of color images composed of pixels by 3000 × 2000pixel; if the amount of data for each pixel using 8bit said, then the need to use three digital images: 3000 × 2000 × 8 × 3 =
144 × 106 bit, but only one image is equivalent to a frame in the video, assuming 25 frames per second playback rate, then the 1s of data is about 25MB, a 600 MB hard disk can only store 24s of moving images. Therefore, the management of video data one
of the keys is the video data compression encoding and decoding. In addition to this, the video data as an expression of information, media, diversity in content, such as may be contained in the video refers to semantic content, but also may refer to the video
contained in the color, texture, object movement, object relationship between the camera operations, object size, shape and so on. Video data also has to explain the diversity and ambiguity, different people on the same section of video may have
different interpretations. Video Retrieval is from a large number of video data needed to find the video clips. The traditional video retrieval is mainly fast-forward and
rewind through methods such as manual search, unable to meet the requirements of multimedia database. Early commercial multimedia databases, such as the VOD system can only provide the keyword-based search or category browsing, retrieval unit
confined to a movie or the whole game, for a smaller video clips, such as a scene or a shot of the search, can only rely on traditional fast-forward, rewind and other means.
And users often want to give examples or as long as the characterization, the system will automatically find the video clips. Video data contains an extremely rich semantic content, but the theoretical level, the video is two-dimensional pixel array of time series,
with the semantic content is not directly related to the.
Therefore, to achieve content-based video retrieval, we must break through the
traditional based on one or more keywords (or attributes) indexing and retrieval based on the limitations of the expression directly to video content analysis, feature extraction
and semantics, and use these content-based feature indexing. Therefore, content-based
retrieval refers to the object, according to media and media content of the semantic and contextual retrieval.
2, content-based video analysis
Video data model is characterized by: each video data is a complex entity, relationship is not there at all between the video data blocks, but in the video data
within the block. So, first of all make video data decomposition, separation of structure and hierarchy. Then analyzes the structure of each object, extract the characteristics of each object, and store these attributes, allowing users to video content based retrieval.
Content-based video analysis, is defined as a specific purpose, from the input video extract relevant information about the content of all the process. In order to achieve the content-based video shot retrieval, video analysis of the basic process includes shot boundary detection, video data of the low-level features automatic indexing and video
clustering. Shot boundary detection via video frame comparison, the video is divided into the basic composition of the unit - lens; automatic indexing video data including
key frame comparison, static characteristics and motion feature extraction, etc.; the basis of these features can be video clustering .
Video analysis of the basic process is as follows:
Third, at home and abroad on the topic Research
1. Shear detection and shot segmentation
Video camera is a basic unit, which consists of a set time frame attached to the composition. Lens testing is a separate video stream cut into a lens. Need to determine
the time when the lens border, or to detect changes in the lens or switching office.
Common video programming switch in the lens can be divided into two kinds: One is the direct switch, known as shear; the other is the optical switch is a corresponding
gradual scene change, known as the gradient.
Reposted elsewhere in the paper for free download http://
Detection of these two switch the order of a strategy is to detect them: first, shear inspection after inspection gradient. The input video stream is the original video stream or a compressed video stream. The former use of neighborhood average, the latter extract the DC component, can be a video stream to be detected. Camera switching, the video data will undergo a series of changes, showing a sudden increase in the color difference between the new and old away from the edge of the object shape changes and movements are not continuous and so on. The purpose of shot boundary detection is to find the law of these changes. In general, the same lens in a smaller difference between the various frames, and different inter-frame lens quite different.
2. Key Frame Extraction
The key-frame camera is the lens to reflect the main information content of the
frame. Will be detected by the camera after the shot can be extracted for each key frame, and use the key-frame to express succinctly the lens. This is because each shot is taken under the same scene, the same scene in each frame there is considerable duplication of information, the key-frame that is reflected in the main information
content of the lens frame images, the general use of a lens extracted out of one or several frames to represent. In addition, the lens with the key frame that makes the image-based technology can be used for video shot retrieval.
3. More well-known image / video retrieval system
QBIC: is the IBM-developed commercial image retrieval system, which supports: Based on the sample images of the query, the user configuration sketches, the user
draw graphics, the user selects desired texture and color.
VIRAGE: yes VIRAGE INC graphics developed by content-based search engine,
similar to the QBIC, VIRAGE support based on color, color layout, texture, structure, and other visual information retrieval, support the combination of several atomic queries query, the user can according to their own will adjust the weight of a query.
PHOTOBOOK: MIT Media Lab developed a set of interactive tools for browsing and retrieval, it implements the shape, texture and facial feature extraction and retrieval.
VISUALSEEK and WEBSEEK: VISUALSEEK is a visual feature search engines, WEBSEEK-oriented WEB text / image search engine, developed by the COLUMBIA University.
NETRA: UCSB for ALEXANDRA Digital Library project has developed a prototype system, it uses color, texture, shape and segmentation of the image area after the relationship between the visual characteristics of the airspace.
MARS: is the University of Illinois at URBANA CHAMPAIGN developed.
BLOBWORLD: yes UC BERKELEY development. Will the original image is converted to a group of local color and texture-related, allowing users to view images of
internal representation and query results, allowing users to visually improve the search
Content-based Video Retrieval System key technologies include: shot cut detection and segmentation; key frames and representative frame extraction; video data indexing; video data representation; user's query retrieval.
 Liu Zheng-kai, TANG Xiaoou. Video Retrieval Methods for shot segmentation [J]. Computer Engineering and Applications, 2002, (23)
 to Germany, Lu Ma. Content-based video structure model [J]. Railway Society, 2000, (4) reposted elsewhere in the paper for free download http://