3D face reconstruction based on progressive cascade regression

Abstract

In order to better learn the distributions of 2D and 3D faces and the mapping between them with limited training samples, a new 3D face reconstruction method based on progressive cascade regression is proposed. Firstly, it learns the mapping between 2D and 3D facial landmarks to estimate the initial 3D facial landmarks with a coupled space learning method. Secondly, a deformed space is constructed with the difference between the estimated initial landmarks and the ground truth of training samples; and more accurate 3D facial landmarks are reconstructed by modifying the initial 3D ones with shape compensations which are calculated by minimizing an objective function. Finally, the realistic 3D faces are reconstructed by a method that is based on a simple sparse regulation and shape deformation. The results on BJUT 3D face database demonstrate the effectiveness of the proposed method. In addition, compared with some typical methods, the method can get better subjective and objective results, especially in details.