3D Face Reconstruction and Dynamic Feature Extraction for Pose-Invariant Face Recognition
Xiaohu Shao, Xi Zhou, Cheng Cheng, Tony Xiao Han · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2013
In this paper, we present a pose invariant face recognition framework leveraged on 3D face reconstruction and dynamic feature extraction.First, we synthesize the virtual frontal face from the probe face based on 3D face reconstruction.In the initialization of reconstruction, a treestructured model is applied to detect landmark points from a 2D image and a hierarchical gaussianization (HG) based method is used for pose estimation.Second, in the recognition step, we present a dynamic feature extraction method to improve the recognizer, which measure the similarity between the synthesized the virtual face and the probe face.Recognition experiments are carried on the Multi-PIE, and CIGIT Face databases.The experimental results show that our system significantly improves the accuracy of face recognition, especially for the faces with extreme poses.