iPDO: An Effective Feature Depth Estimation Method for 3D Face Reconstruction

We present a 3D face modeling approach under uncontrolled conditions. In the heart of this work is an efficient and accurate facial landmark depth estimation algorithm. The objective function is formulated by similarity transformation among face images. In this method, pose parameters and depth values are optimized iteratively. The estimated 3D landmarks then are taken as control points to deform a generic 3D face shape into a specific face shape. Test results on synthesized images show that the proposed methods can obtain landmarks depth both effectively and efficiently. Whats' more, the 3D faces generated from real-world photos are rather realistic based on a set of landmarks.