Reaearch On Content-based Video Shot Detection And Classification

In recent years, the rapid development of electronics industry,multimedia technology and network technology make video resources in the web have more and more users.There are plenty kinds of video resources,such as video on demand,digital television,digital libraries,video conferencing,distance education and so on.They have some features,such as large volume of data, complex multi-dimensional structure and rich content.Therefore, how to find the desired video information rapidly among these vast amounts of unstructured video data resources has become an urgent problem. Video resource management and retrieval research have also become a hot topic.We have developed a variety of video retrieval systems already.The paper presented a shot classification-based video retrieval system by analysing the video contents.In order to automatically manage the video resources following the thread of content-based retrieval.The paper made more in-depth researchment of the video structure,content analysis and retrieval techniques.Major work and innovation are as follows:First,we raised a content-based video shot classification system by observing a large number of video resources.we made a detailed understanding of various shot language and the use of the shots in different scene.We developed a classification principle and then we classfied the shots according to different classification criterias.We also analysed the visual characteristics of various types of shots.Second, this paper presented a multi-classification strategy for video shot boundary detection after we researched several shot detection methods.By extracting a variety of effective features from the video, we used SVM to detect different shot changes.At the same time,We achieved the classification of shots on different change styles.Third, in the shot classification,we realized the recognition of the video shot motion pattern by extracting the dynamic features of the video shots-optical flow analysis.Then we took an appropriate number of key frames according to different move patterns,and made video summary of the video frames.We also realized the auto recognization of taking shots,creative shot,animation shots,demo shots by extracting the static characteristics of the key frames.Fourth,we designed and realized a shot classification-based video retrieval system.Also we made an index database of the video shot.We carried out experiments on video resources and achieced good results.The experimental plat of the video shot detection,the video shot analysis system and the shot classification-based video retrieval system are all based on Windows XP operating system and reliaze using Visual C + +6.0,with the DirectSDK package.The research results show that the management of the video resources in different levels not only achieved a fast and efficient retrieval of resources,but also provided clues between the underlying characteristics and high-level semantic features of the video.In addition, We can get the characteristics of certain types of video through analysing the shots characteristics ,so as to realize the automatic classification of different types of video,and establish links between low-level features and high-level semantic.