3-dimensional face pose and shape estimation based on relaxed model fitting optimization

Three-dimensional face pose and shape acquisition has become an important part of 3D face reconstruction, 3D face tracking, and other related applications. Many approaches have been presented to estimate the 3D face shape from 2D images. Some of them are parameterized model based approaches so it is somewhat convenient to handle model shape and pose via numerical parameters to align its vertices on the target face image instead of explicitly adjusting each vertex. However, the important factors that have usually been concerned and needed to improve are the computation complexity and also the process reliability. In this paper, we propose an efficient way derived from the steepest descent to more accurately estimate face pose and shape from a 2D image as it can escape from the local minima.