一种基于动态时间规划的视频特征检索改进算法  

An improved video retrieval algorithm based on dynamic time warping

作　　者：王一梅[1] 贾克斌[1] 庄新月[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100022

出　　处：《高技术通讯》2007年第5期464-469,共6页Chinese High Technology Letters

基　　金：863计划（2004AA103960）资助项目,北京市教委科技发展重点基金（ZK200410005004）、北京市拔尖创新人才计划（000）资助项目.

摘　　要：针对基于“视频指纹”特征的视频检索算法在实际应用中存在视频亮度整体漂移、突变干扰以及视频再编辑的问题,提出了一种基于动态时间规划的视频特征检索改进算法。该算法在原“视频指纹”算法的基础上采用了新的视频检索策略：首先,对视频帧进行区域分割,将“视频指纹”由单分量扩展到多分量;然后,采用基于粒度的相似性比较算法,用比对每帧视频指纹的变化,取代比较视频指纹本身;最后,采用改进的动态时间规划算法进行视频特征匹配,定位目标视频。在算法评估中,建立了由74段广告视频和154段再编辑视频组成的测试集。通过试验证明,改进算法可以有效应对视频检索过程中的亮度整体漂移、突变干扰和视频再编辑问题,检索算法具有很好的鲁棒性。An improved video retrieval algorithm based on dynamic time warping was proposed to solve the practical problems of the original video retrieral algorithm based on video fringerprints, such as offset of color or brightness, abrupt intense disturbance and video re-editing. Video frames were divided into segments, then the new algorithm compared the change of video fingerprints instead of comparing video fingerprints directly and used a modified dynamic time warping algorithm to cope with video re-editing. To evaluate the performance of the new algorithm, a large video dataset consist of 74 TV programs and 154 re-edited programs were set up. Experiment results demonstrated the effectiveness and robusmess of the algorithm.

关 键 词：视频检索 视频指纹 动态时间规划算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]