中文摘要：

针对传统的三维人脸识别算法成本较高且不能很好地处理带有光照、表情等变化人脸识别的问题,设计了低分辨率Kinect传感器采集三维点云的鲁棒人脸识别系统。首先,通过鼻尖检测、人脸剪裁、姿势校正、对称填充及平滑采样得到规范的纹理图像;然后,在纹理图像上运用判别色彩空间变换,从而最大化类与类之间的分离性;最后,利用多模态稀疏编码有效地重建误差以得到查询图像与训练集之间的相似度,并利用Z-得分技术完成最终的人脸识别。在通用人脸数据库CurtinFaces、PIE及AR上的识别率可高达96.7%,实验结果表明,相比其它几种人脸识别算法,本文算法取得了更好的识别效果。

英文摘要：

Traditional three dimensional face recognition algorithms can not deal with robust face recognition but with high costs, a robust face recognition system based on Kinect sensor with low resolution collecting 3D point cloud is designed. Firstly, standardized texture images are got by tip detecting, face cutting, posture correcting, symmetry filling and smooth sampling. Then, discriminant color space transform is used on texture images to maximize separabilities between classes. Finally, multi-modal sparse coding is used to reconstruct errors so as to getting similarities between the query image and total training set, and Z-scoring technique is used to recognize face. Recognition accuracy of proposed algorithm can achieve 96.7% on common face databases CurtinFaces, PIE and AR. Experimental results show that proposed algorithm has higher recognition accuracy and better recognition efficiency than several advanced face recognition algorithms.