Improving Bag-of-visual-Words model with spatial-temporal correlation for video retrieval

摘要

Most of the state-of-art approaches to Query-by-Example (QBE) video retrieval are based on the Bag-of-visual-Words (BovW) representation of visual content. It, however, ig- nores the spatial-temporal information, which is important for similarity measurement between videos. Direct incorpo- ration of such information into the video data representa- tion for a large scale data set is computationally expensive in terms of storage and similarity measurement. It is also static regardless of the change of discriminative power of vi- sual words with respect to di↵erent queries. To tackle these limitations, in this paper, we propose to discover Spatial- Temporal Correlations (STC) imposed by the query exam- ple to improve the BovW model for video retrieval. The STC, in terms of spatial proximity and relative motion co- herence between di↵erent visual words, is crucial to identify the discriminative power of the visual words. We develop a novel technique to emphasize the most discriminative visual words for similarity measurement, and incorporate this STC-based approach into the standard inverted index archi- tecture. Our approach is evaluated on the TRECVID2002 and CC WEB VIDEO datasets for two typical QBE video retrieval tasks respectively. The experimental results demon- strate that it substantially improves the BovW model as well as a state of the art method that also utilizes spatial- temporal information for QBE video retrieval.