基于 KINECT 传感器采集三维点云的高效鲁棒人脸识别

摘要

For the problem that traditional 3D face recognition algorithms cost high but still can not deal with robust face recognition,we propose an efficient and robust face recognition algorithm which is based on collecting 3D point cloud by low resolution KINECT sensor.First, for the problem that the data collected by KINECT sensor has big noises,we present the standardised pretreatment process,it is to get the standardised texture images by nose tip detecting,face image cropping,posture correcting,symmetry filling and smooth sampling.Then,we run discriminant colour space transform on texture images so as to maximise the separability between the classes.At last,we use multi-modal sparse coding to effectually reconstruct the errors for getting the similarity between the querying image and the training set,and use Z-scoring technique to complete final face recognition.The efficiency and robustness of the proposed algorithm have been verified by experiments on common face database CurtinFaces.Experimental results show that the proposed algorithm achieves higher recognition accuracy and better robustness than several other relatively advanced robust face recognition algorithms.%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鲁棒性人脸识别的问题，提出一种基于低分辨率 KINECT 传感器采集三维点云的高效鲁棒人脸识别算法。首先，针对 KINECT 传感器采集到的数据噪声大的问题，提出规范化预处理过程，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 上的实验验证了算法的高效性及鲁棒性。实验结果表明，相比其他几种较为先进的鲁棒人脸识别算法，该算法取得了更高的识别率及鲁棒性。