| With the dramatic increase of digital video and the appearance of various multi-media applications, such as digital library, video-on-demand, remote instructions and so on, how to quickly locate required information from huge amount of videos becomes more and more important. Due to their well-known limited description ability, strong objectivity, and requirement for manual labeling, the traditional keyboard-description-based retrieval approaches can no longer meet the retrieval demand of users. Therefore, content-based Video Retrieval (CBVR) became a hot topic in current research. Furthermore, because video retrieval is normally applied in web, research on content-based video retrieval under web circumstances is of essential value.This paper would mainly focus on the following three aspects:(1) We designed a universal user retrieval interface by integrating several retrieval ways together, including video browsing, example retrieval, feature combination, semantic object, text remark. It overcomes one of the disadvantages of single retrieval approach that the description of retrieval requirements is not clear. It can help users to describe their video retrieval requirements more easily and clearly in an interactive way to get more satisfactory retrieval results. In addition, we would also present a new relevance feedback approach by integrating weight-based adjusting relevance feedback approach and the recording of feedback log together. With this new method, it can reduce number of user feedbacks and in this way to improve the efficiency of retrieval.(2) Considering the function of the user retrieval interface, we discuss the critical techniques in the realization of content-based Video retrieval. Our approach can be described as following: Choosing easy and feasible methods of spot detection, key-frame extraction and scene detection for the video retrieval system based on comparison and analysis; Standardizing the selection of key-frame feature referring to... |
PDF Full Text Request