| Multi-media technology developed In the late 80's, which has become a global hot spots of technology research and product development, especially to the video retrieal and mining, which have broad application fields and development prospects. Digital Video has focued on the most important development area of information industry in our country. As the rapid development of digital TV and broadband networks, video on demand, interactive TV, video websites will become more and more extensive, these applications will ultimately lead to a large number of sharp increase of video data. However, due to the unstructured digital video data, data volume and other features such as redundant data, existing or future video data arising from a large number of digital video can not access to utiliz. a large number of valuable data can only become the meaningless accumulation, so to research how to use of a large number of video data, mining meaningfull video information for our decision from a large number of video data become a important and urgent problem.Video retrieval and video mining are coming from this hot demand, it combines the digital image processing, digital video processing, multimedia technology, database technology, artificial intelligence technology, pattern recognition technology, probability statistics, fuzzy mathematics and other areas, using of bottom and in essential features of video images for retrieval to overcome the deficiency of traditional text based retrieval system. The paper discussed key technologies of video retrieval and video mining in compressed domain. A serial of problom such as video shot detection, video key frame extraction, video summary extraction and video descriptor extraction technologies are detailed, and achieved following results.1. To the video shot detection, a highly efficient video shot detection algorithm is proposed based on I-frame in compressed domain. After the in-depth analysis of MPEG-2 international standard, we can conclude the video data encomposed I frames, P frames and B frames, and I frame is the major carriers to the video data, and it emboy DCT coefficients, after the preprocessing, DC coefficients can be achieved from DCT coefficients. the using of DC coefficients in detection can significantly reduced the volume of video data. And on the other hand, it can overcome the computing complex of traditional non-compressed domain, which significantly improved the efficiency of the video shot detection.2. To the Key Frame Extraction, a new algorithm based on RS is proposed in compressed domain, firstly, the algorithm extract the I frame and its DCT components and DC components. Then construct the information system using DC coefficients, after using the attributes reduce theory of RS, the I frames that can not reduce are achieved and its can represent the key frame of video. On the one hand, the algorithm introduce the RS theory into key-frame extraction at the first time, and it can scientific and efficient access to key-frame out of video, overcoming the deficiency of unscientificity in traditional key-frame extraction algorithm, on the other hand, the algorithm directly processed in compressed domain to avoid large amount of computing.3. To the video summary extraction, the paper proposed a automatic video summary extraction algorithm based on RS in compressed domain. As to a video shot or video lens, an important content must need more video frame that has little change to embody, that is the more important content, the more frames need to repeat. According to this principal, the proposed algorithm reduce the I frame according to the RS theory. Then, sorting the reduced I frame according to the number in each reduced I frame.4. To the video descriptor extraction, the paper proposed a space-time color descriptor extraction algorithm. According to the MPEG international stardand, the algorithm extract DCT coefficient and DC coefficient from video stream, constructing the information system using DC coefficients, after using the reduced theory of RS, achieved the core of information system. Since it represent the video frame that can not reduce, so combined the core frame can achieve effective video descriptor. The order of video signify the time attributes and the DC coefficient contained the space attributes, so it can be seen an effective video descriptor.
|
PDF Full Text Request