蔡川丽, 张建平, 张彦博. 一种抗姿态与表情变化的三维人脸识别方法[J]. 应用光学, 2018, 39(4): 491-499. DOI: 10.5768/JAO201839.0402002
引用本文: 蔡川丽, 张建平, 张彦博. 一种抗姿态与表情变化的三维人脸识别方法[J]. 应用光学, 2018, 39(4): 491-499. DOI: 10.5768/JAO201839.0402002
Cai Chuanli, Zhang Jianping, Zhang Yanbo. 3D face recognition method against gesture and expression change[J]. Journal of Applied Optics, 2018, 39(4): 491-499. DOI: 10.5768/JAO201839.0402002
Citation: Cai Chuanli, Zhang Jianping, Zhang Yanbo. 3D face recognition method against gesture and expression change[J]. Journal of Applied Optics, 2018, 39(4): 491-499. DOI: 10.5768/JAO201839.0402002

一种抗姿态与表情变化的三维人脸识别方法

3D face recognition method against gesture and expression change

  • 摘要: 为了提高人脸在姿态和表情变化下的识别率,结合局部平面距离(DLP)对曲面局部凹凸性优良的判断能力,提出了一种采用人脸的等距不变表示形式来匹配的人脸识别方法。首先,对深度摄像头采集到的深度图像进行距离约束、位置约束、转换等操作,得到干净完整的三维人脸,利用三维人脸上每一点DLP值确定鼻尖点,利用聚类的思想确定鼻根点;其次,采用改进的快速推进算法计算人脸的测地距矩阵,设置阈值并切割出有效的人脸区域;最后,计算有效的人脸区域的高阶矩特征,作为人脸的特征向量进行匹配。实验结果表明,对于不同的数据库,本文算法的识别率接近97%;将本文算法与基于轮廓线特征的人脸识别算法以及基于Gabor特征的人脸识别算法进行比较,其识别率分别提高了14.1%和8.3%,同时有着较高的运算效率。

     

    Abstract: In order to improve the facial recognition rate under the change of posture and expression, combined with the ability of local plane distance (DLP) to judge the convexity of local curved surface, a face recognition method based on face equidistant invariant representation was proposed. Firstly, several operations such as distance constraint, location constraint and transformation were conducted on the depth image captured by the deep camera to get the clean and complete 3D face; then the nose tip was determined by the DLP value of every point on the 3D face, and the nasal root was determined by the clustering idea; secondly, the improved fast propulsion algorithm was used to calculate the geodesic distance matrix of face, then the threshold value was set and the effective face area was cut out; finally, the high-order moment feature of the effective face area was calculated as the feature vector of face for matching. The experimental results show that the recognition rate of this algorithm is close to 97% for different databases. Compared with the face recognition algorithms based on contour features and Gabor features, the recognition rate of this algorithm is increased by 14.1% and 8.3%, respectively, while having a high computing efficiency.

     

/

返回文章
返回