Experimenting VIREO-374: Bag-of-Visual-Words and Visual-Based Ontology for Semantic Video Indexing and search.
Chong‐Wah Ngo, Yu–Gang Jiang, Xiao-Yong Wei, Feng Wang, Wan‐Lei Zhao, Hung‐Khoon Tan, Xiao Ying Wu · 2007
In this paper, we present our approaches and results of high-level feature extraction and automatic video search in TRECVID-2007. In high-level feature extraction, our main focus is to explore the upper limit of bag-of-visual-words (BoW) approach based upon local appearance features. We study and evaluate several factors which could impact the performance of BoW. By considering these important factors, we show that a local feature only system already yields top performance (MAP = 0.0935). This conclusion is similar to our recent experiment of VIREO-374 on TRECVID-2006 dataset [1], except that the improvement, when incorporating with other features, is marginal. Description of our submitted runs:- CityU-HK1: linear weighted fusion of 4 SVM classifiers using BoW, edge histogram, grid based color moment and wavelet texture.- CityU-HK2: average fusion of 5 SVM classifiers using BoW, spatial layout of keypoints, edge histogram, grid based color moment and wavelet texture.- CityU-HK3: average fusion of 4 SVM classifiers using BoW, edge histogram, grid based color moment and wavelet texture.- CityU-HK4: Bag-of-visual-words (BoW).- CityU-HK5: average fusion of 3 baseline classifiers using edge histogram, grid based color moment and wavelet texture.- CityU-HK6: average fusion of 2 baseline classifiers using grid based color moment and wavelet texture.