LSDA Solution Schemes for Modelless 3D Head Pose Estimation

Locality Sensitive Discriminant Analysis (LSDA) is a recent linear manifold learning method used in pattern recognition and computer vision. Whenever LSDA is used for face image analysis, it suffers from a number of problems including the Small Sample Size (SSS) problem, and that its classification performance seems to be heavily influenced by its parameters. In this paper, we propose two novel solution schemes for LSDA. The first solution is a novel parameterless approach. The second solution is an exponential LSDA which can solve the SSS problem, in the sense that there is no need to perform the pre-stage of dimensionality reduction. The proposed solution schemes have been applied to the problem of modelless coarse 3D head pose estimation. They were tested on two databases FacePix and Pointing’04. They were conveniently compared with other state-of-the-art linear techniques. The experimental results confirm that our methods can give better results than the existing ones.