Improving dynamic texture recognition with constraint subspace learning

抄録

We present a new framework for recognizing dynamic textures modeled by linear dynamic systems (LDS). The new framework improves previous methods by considering between-class subspace differences. Specifically, common part among different class subspaces is removed by projecting model subspaces onto a constraint subspace before recognition. This operation increases the separability among different classes of dynamic textures. The effectiveness of the new framework is shown by experimental results.