摘要
针对金字塔匹配下的视频检索系统中基础特征用矢量量化方法表示不够精确的问题,结合稀疏编码方法进行视频检索。视频的基础特征通过稀疏编码表示后,用金字塔方法进行多次匹配,将多次匹配结果线性合并,作为修正后的相似性度量结果。通过对UCF50的检索实验表明,该方法能显著提高检索的准确率。
This paper proposes an extended video indexing approach combined with the sparse coding method, in order to improve the accuracy for feature vector quantizing within pyramid matching, which has been widely applied in CBVR systems. Essential features of video sequence, extracted from the adjacent two frames, are coded by the sparse coding method previously, and then registered by the pyramid matching for several times, in order to produce slight different alignments. After that, these alignments between two frames are linearly combined to be the corrected similarity measurement. The experiment on UCF 50 shows the proposed method can greatly increase the accuracy of video retrieval task.
出处
《计算机工程与应用》
CSCD
2013年第21期191-194,共4页
Computer Engineering and Applications
基金
国家自然科学基金(No.61075019)
关键词
视频检索
金字塔匹配
稀疏编码
video retrieval
pyramid matching
sparse coding
