3D Facial Expression Recognition Based on Combination of Local Features and Globe Information

This paper presents a novel 3D facial expression recognition algorithm using Local Binary Patterns (LBP) under expression variations, which has been extensively adopted for facial analysis. First, to preserve the main information and remove noises which will affect the discrimination, BDPCA reconstruction and Shape Index is utilized to depict the human face accurately. Then the LBP framework for face representation is introduced. The facial local features and global information are all extracted to encode by LBP to reduce the influence of expression changes. The faces are represented by taking advantage of the statistical histograms that fuse the facial detailed information of each face region and the whole face. And the Chi-square is used as matching strategy to address the recognition task. Finally, the proposed algorithm was tested on Bosphorus database, achieving promising recognition rate. It proves that our algorithm is robust to expression variations and is feasible to apply on 3D facial expression recognition.