Voting Based Video Classification Using Clustering and Learning

In this paper, we propose video classification using linear manifolds (affine subspaces). In our method, we represent videos belonging to a same class by several linear manifolds using k-varieties clustering. When a test video is given, each frame of it votes for the class to which its nearest linear manifold belongs. According to this voting, the test video is classified into the class that achieves the majority votes. For improving accuracy, a way of adopting generalized learning vector quantization for our video classification is also presented. The performance of our video classification is verified with experiments on short videos downloaded from Web.