Two dimensional partitioned sparse representation for head pose estimation

Although classical sparse representation is capable to solve appearance based classification problems such as face recognition, it is problematic that images need to be converted to column vectors before subsequent processing which makes the computation expensive due to the huge dimension. From the human vision perspective, it is reasonable to observe image in form of matrix rather than vector. To reduce the computational complexity, the idea to partition the images is introduced as well. We combine partition processing with two dimensional sparse representation together to propose 2DPSRC (2D Partitioned Sparse Representation Classifier) considering the property of head pose estimation problem. It can greatly improve the estimation accuracy and enhance the efficiency of the computational process involved pursuit of ℓ1-norm minimization. Finally, experiments on Pointing'04 and Oriental Face Database show the effectiveness and robustness of our proposed method.