The method based on Bag-of-visual-Words (BoW) deriving from local keypoints has recently appeared promising for video annotation. Spatial partition scheme has critical impact to the performance of BoW method. In this paper, we propose a new adaptive annular spatial partition scheme. The proposed scheme firstly determines the centroid of partition according to the distribution of keypoints. And then the image is partitioned into several annular regions. In the end, BoW histograms are computed according to the annular regions, which are utilized to train SVM classifiers. A systematic performance study on TRECVID 2006 corpus containing 20 semantic concepts shows that the proposed scheme is more effective than other popular spatial layout partition schemes such as 2 × 2 grid scheme.
[1]
Paul Over,et al.
TREC video retrieval evaluation TRECVID
,
2008
.
[2]
Dong Wang,et al.
THU and ICRC at TRECVID 2007
,
2007,
TRECVID.
[3]
Chong-Wah Ngo,et al.
Towards optimal bag-of-features for object categorization and semantic video retrieval
,
2007,
CIVR '07.
[4]
G LoweDavid,et al.
Distinctive Image Features from Scale-Invariant Keypoints
,
2004
.
[5]
Chong-Wah Ngo,et al.
Columbia University/VIREO-CityU/IRIT TRECVID2008 High-Level Feature Extraction and Interactive Video Search
,
2008,
TRECVID.