An accurate hierarchical binocular stereo approach to reconstruct 3D facial model

A new hierarchical dense matching method using Daisy features is proposed to reconstruct accurate 3D face with binocular camera system. The Daisy operator has high precision and high speed advantages, but the huge memory consumption limited its application in the high resolution image matching algorithms. A divide and conquer strategy is proposed to deal with huge memory consumption issue in high resolution image matching with DAISY features. Firstly, An initial disparity map has been acquired with a hierarchical matching. And then a modified disparity refinement has been executed to get a high accuracy point clouds. At last, high accurate face model can be reconstructed using Poisson Surface Reconstruction (PSR) method. The only inputs of our method is a pair of calibrated face images captured in standard illumination. Compared with other face reconstruction methods, the experimental results shows our algorithm is simple to implement and highly accurate.