A Feature Sequence Kernel for Video Concept Classification

Kernel methods such as Support Vector Machines are widely applied to classification problems, including concept detection in video. Nonetheless issues like modeling specific distance functions of feature descriptors or the temporal sequence of features in the kernel have received comparatively little attention in multimedia research. We review work on kernels for commonly used MPEG-7 visual features and propose a kernel for matching temporal sequences of these features. The sequence kernel is based on ideas from string matching, but does not require discretization of the input feature vectors and deals with partial matches and gaps. Evaluation on the TRECVID 2007 high-level feature extraction data set shows that the sequence kernel clearly
 outperforms the radial basis function (RBF) kernel and the MPEG-7 visual feature kernels using only single key frames.