Journal of Software:2003.14(8):1409-1417

An Approach for Video Retrieval by Video Clip
PENG Yu-Xin,NGO Chong-Wah,DONG Qing-Jie,GUO Zong-Ming,XIAO Jian-Guo
Chart / table
Similar Articles
Article :Browse 3672   Download 4132
Received:November 25, 2002    Revised:March 20, 2003
> 中文摘要: 视频片段检索是基于内容的视频检索的主要方式,它需要解决两个问题:(1) 从视频库里自动分割出与查询片段相似的多个片段;(2) 按照相似度从高到低排列这些相似片段.首次尝试运用图论的匹配理论来解决这两个问题.针对问题(1),把检索过程分为两个阶段:镜头检索和片段检索.在镜头检索阶段,利用相机运动信息,一个变化较大的镜头被划分为几个内容一致的子镜头,两个镜头的相似性通过对应子镜头的相似性计算得到;在片段检索阶段,通过考察相似镜头的连续性初步得到一个个相似片段,再运用最大匹配的Hungarian算法来确定真正的相似片段.针对问题(2),考虑了片段相似性判断的视觉、粒度、顺序和干扰因子,提出用最优匹配的Kuhn-Munkres算法和动态规划算法相结合,来解决片段相似度的度量问题.实验对比结果表明,所提出的方法在片段检索中可以取得更高的检索精度和更快的检索速度.
Abstract:Video clip retrieval plays a critical role in the content-based video retrieval. Two major concerns in this issue are: (1) automatic segmentation and retrieval of similar video clips from video database; (2) similarity ranking of similar video clips. In this paper, motivated by the maximal matching and optimal matching in graph theory, a novel approach is proposed for video clip retrieval based on matching theory. To tackle the clip segmentation and retrieval, the retrieval process is divided into two phases: shot-based retrieval and clip-based retrieval. In shot-based retrieval, a shot is temporally partitioned into several sub-shots based on motion content. The similarity among shots is measured according to the color content of sub-shots. In clip-based retrieval, candidates of similar video clips are selected by modeling the continuity of similar shots. Maximal matching based on Hungarian algorithm is then adopted to obtain the final similar video clips. To rank the similarity of the selected video clips, four different factors: visual similarity, granularity, interference and temporal order of shots are taken into consideration. These factors are modeled by optimal matching based on Kuhn-Munkres algorithm and dynamic programming. Experimental results indicate that the proposed approach is effective and efficient in retrieving and ranking similar video clips.
文章编号:     中图分类号:    文献标志码:
Foundation items:
Reference text:

彭宇新,Ngo Chong-Wah,董庆杰,郭宗明,肖建国.一种通过视频片段进行视频检索的方法.软件学报,2003,14(8):1409-1417

PENG Yu-Xin,NGO Chong-Wah,DONG Qing-Jie,GUO Zong-Ming,XIAO Jian-Guo.An Approach for Video Retrieval by Video Clip.Journal of Software,2003,14(8):1409-1417