Video segment retrieval based on affine hulls

In this paper, we propose a novel video segment retrieval algorithm, which is based on the affine hulls [15]. As we all know, shot is the basic unit of retrieval, so we approximate each of the shots with an affine hull of the frames within it, after that each shot is represented by a mean vector and a series of basis vectors. Therefore, we can get the distances between the query frame and each of the shots, if we sort them in ascending order, then the first few indicate the result we get. This algorithm is compared to the K-means clustering algorithm for video shot representation and retrieval. In the experiments, three video sequences about agriculture were used to evaluate the performances of the different retrieval algorithms. Experimental results show that our proposed affine hulls based algorithm outperforms the K-means clustering algorithm for video retrieval.