An efficient approach to content-based object retrieval in videos

Traditional methods for the content-based object retrieval suffers high space and time consumption. In this paper, we propose a novel method for the efficient object retrieval in videos. First, we compress the feature descriptors by tracking the feature points between consecutive frames and encoding the tracked feature points. The encoding procedures are performed by applying the motion prediction in video codec. Second, we propose a new algorithm to locate the objects by searching for the high-density positions of the related feature points in frames. To improve the speed, we count the corresponding words of feature points within the query target and calculate their spatial distributions. Experimental results show that the proposed method outperforms the previous methods.