Example-guided 3D Human Face Reconstruction from Sparse Landmarks

Abstract

This paper presents an example-guided facial reconstruction method for creating a 3D refined human face model from sparse landmarks, with a given example dataset. It is challenging to rapidly and accurately generate a high-precision 3D face model from raw scan data with a simple setup and processing. The main roadblock is that existing 3D face databases are far from adequate to describe the full variability of faces. To address these problems, we analyse the characteristics of examples from a relatively small sample set for more reliable reconstruction knowledge, as well as a new landmark marking method to simplify the description of human face shape. Principal component analysis is used to extract the feature patterns from samples and simplify data representation. Then the 3D face model and the landmarks are correlated via a mapping matrix. An effective mapping algorithm is devised to learn the transformation relation from landmarks to 3D face shapes. Compared with existing methods, the proposed method can generate a high-precision 3D face model from sparse landmarks more accurately. The application and extensive experimental evaluations on the Chinese craniofacial database and FaceWarehouse database show that our method can achieve high accuracy, effectiveness and robustness in 3D face reconstruction.