The Research Of The Shot Boundaries Detection Based On Textual Features

With the rapid development of computer technology and network communication technology, multimedia data, such as video, has gradually became the main form of information media in information processing. While the continuous expansion of the network and the continuous of development of the access technology provide vast space for broadcasting the video and others multimedia. With the multimedia data, especially the video data, is becoming more and more. The operation to the video data is not only limited in storage and transmission, the video retrieval application have been accepted by more and more people. But the inefficiencies and limitations of the traditional retrieval methods application makes it have a good demand for a new retrieval method which can deal with the video data. Thus the content-based video retrieval appears.The major research of this paper is the key technologies of content-based video retrieval—the shot boundaries detection.First this paper Summarize and introduce the existing methods of shot boundaries detection, and analyze the main problems of the existing methods of shot boundaries detection, then propose the textual features to shot boundaries detection. Secondly this paper Summarize and introduce the existing methods of extracting various kinds of textual features—pixel-based domain,based on spectrum and fractal, through analysis and comparison, we decide to approach fractal dimension as the textual features. Then I propose an improved key-frame extraction method based on unsupervised clustering. After that this paper establishes a video shot boundaries detection system based on textual features. This system mainly include three parts—image pre-processing,extracting image features,shot boundaries detection and key-frame extraction. Image pre-processing mainly filter and block for every frame in the video clips; the part of extracting image features extract textual features form every frame by DBC method; video shot boundaries detection approach sliding windows and dynamic threshold to extracting shot boundaries of video clips. Finally through emulation, the proposed video shot boundaries detection system is verified, the results verified the proposed system has very good robustness to cut, especially to gradual change, at the same time be able to efficiently extract key-frame based on the different video content.