Robust Regression for Image Alignment via Subspace Recovery Techniques

Abstract

We present a novel method for joint head pose estimation and face alignment via subspace recovery techniques by incorporating an affine transformation. The new algorithm seeks a set of optimal affine transformations to fix the geometric distortions and deal with a variety of adverse effects such as illumination and occlusions, outliers and heavy sparse noises. Our method is also formulated as a convex optimization problem which can be solved by using an augmented Lagrangian multiplier and takes the advantages of Jacobean transformation matrix in transforming the corrupted images. The convergence analysis is shown to prove the effectiveness of the proposed approach. Conducted simulations justify the superiority and effectiveness of the proposed approach as compared with the main state-of-the-art works.