Visual tracking based on direct orthogonal locality preserving projections

This paper proposes a novel algorithm for feature extraction, called Direct Orthogonal Locality Preserving Projections (DOLPP). The proposed algorithm is based on the orthogonal locality preserving projections (OLPP), which aims at finding a set of orthogonal basis vectors to the eigenfunctions of the laplace beltrami operator. The DOLPP can directly compute the orthogonal basis and is expected to have more discriminating power than LPP. Experimental results on video sequences demonstrate the effectiveness of the propose algorithm.