A Method of Object-Based Video Retrieval Using Spatiotemporal-Word Pairs
This paper builds upon previous work on local invariant regions detection and description. We first segment the video into a set of shots, then the local regions are tracked throughout a shot and stable tracks which have the attributes of spatial and time are extracted. In order to take into account the structural information within tracks and motivated by the bag-of-words model, we propose spatiotemporal-word pairs based approach to retrieve frames containing desired objects. The spatiotemporal-word pair is defined as a pair of spatiotemporally adjacent tracks. We devise the method on how to construct spatiotemporal-word pairs for each shot and how to encode the spatiotemporal-word pairs for indexing and retrieval. The proposed method is applied to object retrieval experiments, the performance of our spatiotemporal-word pairs based video retrieval approach is demonstrated.
Feature extraction Track extraction Spatiotemporal-word pairs Object retrieval
Cunna Liu Cui Xie
Department of Information Science and Engineering Ocean University of China Qingdao, China Department of Information Science and EngineeringOcean University of ChinaQingdao, China
国际会议
哈尔滨
英文
197-201
2011-01-18（万方平台首次上网日期，不代表论文的发表时间）