基于三维谱回归的有效三维人脸识别算法

提出一种结合预处理和三维谱回归(3DSR)方法进行三维人脸识别的算法框架,提取有效的判别特征来克服3D人脸图像中一些尚未解决的问题,如噪声、表情和姿态等的影响。首先通过预处理步骤,从输入的人脸图像提取面部区域进行三维数据的匹配,克服大姿态变化的影响并且有效地提高了整个3D人脸识别性能。为处理大的表情变化和数据噪声,引入谱回归的概念,改进的三维谱回归方法可以充分利用局部统计信息的鲁棒性和有效性,并避免通常方法中密集矩阵的特征分解问题,降低了计算复杂度。实验中使用包含大姿态和表情变化的CASIA三维人脸数据库。实验结果显示算法有效、鲁棒、通用性强。

Efficient 3D Face Recognition Algorithm Based on 3D Spectral Regression

The novel preprocessing and 3D Spectral Regression(3D SR) combined framework is used for 3D face recognition to overcome some of the unsolved problems encountered with 3D facial images,such as noises,expressions and poses.We first extract the facial area with registration,and then process it to minimize the effect of large pose variations and effectively improve the total 3D face recognition performance.To solve the large expression variations and data noises,we introduce the concept of spectral regression,which can make good use of the robustness and efficiency of local statistical information and avoid eigen-decomposition of a dense matrix as in popular methods,with a huge saving in computational costs.Our experiments are based on the CASIA 3D face databases which contain large pose and expression variations.Experimental results show good efficiency,robustness and generality of our proposed method.