DCT representations based appearance model for visual tracking

Abstract

In this paper, we present a simple yet effective visual tracking algorithm with an appearance model based on 2D discrete cosine transform (2D-DCT) representations. The DCT has the properties of decorrelation and energy compaction, and is robust against geometry and illumination changes. Hence, it is suitable for appearance modeling and the features of our appearance model are extracted from an optimized low dimensional subspace. In order to adapt to the appearance change caused by environment change or ego motion, we also propose to update the observation appearance model through a nonlinear weighted method. Numerous experiments on some challenging video sequences demonstrated that our algorithm is effective and it considerably outperforms the other methods.