光学 精密工程, 2014, 22 (12): 3391, 网络出版: 2015-01-13   

三维到二维: 人脸本征形状描述图

3D to 2D: Facial intrinsic shape description maps
作者单位
西北工业大学 电子信息学院,陕西 西安 710072
摘要
基于三维数据的人脸识别克服了二维图像数据受光照和姿态影响较大的问题, 但其较高的数据维数约束了它的实际应用。本文针对三维人脸数据的简化描述, 提出了将三维人脸映射至二维表示的本征形状描述图方法。该方法首先基于约束离散保形映射将三维人脸数据微分同构映射到一个局部几何特征保持的二维区域。然后基于人脸曲面几何结构特性和表观特性, 构建二维本征形状描述图, 用于简化对三维人脸数据的描述, 并进行识别验证。基于国际公共人脸数据库FRGC2.0和GavabDB的三维人脸识别实验显示, 本征形状描述图法在姿态变化大于60°时的识别率达到90.6%, 比现有方法高5.9%, 单次匹配时间为7.89 s。该方法将三维人脸识别问题转换为了二维平面图像的识别问题, 有效降低了数据描述的复杂度。得到的结果展示了该方法计算效率高, 且对姿态变化有良好的健壮性。
Abstract
The face recognition based on 3D facial data overcomes the difficulties sensitive to illumination and pose variations in 2D face recognition systems. However, the high computational complexity restricts its practical applications. To simplify the description for 3D face data, a novel strategy to map 3D face data to 2D ones, called 2D intrinsic shape description map, was proposed in this paper. With the strategy, each 3D facial surface was firstly mapped homeomorphically onto a 2D lattice which keeps the local geometrical features based on the constraint discrete conformal. Then ,a 2D intrinsic shape description map was obtained by combining 3D facial geometrical structure and appearance feature for simplifying 3D face representation and for verifying the recognition. The proposed strategy was compared to state-of-the-art 3D face recognition algorithms in the FRGC 2.0 and GavabDB database. The results show that the proposed strategy offers the rank-one rate of 90.6% when the pose change is greater than 60°, higher 5.9% than that of the existing method. Moreover, the single matching time is 7.89 s. As the strategy transforms the 3D face recognition into the 2D image recognition, it effectively reduces the complexity of data description and shows higher computation efficiency and robustness to a pose change.

郭哲, 樊养余, 刘姝, 雷涛, 王毅. 三维到二维: 人脸本征形状描述图[J]. 光学 精密工程, 2014, 22(12): 3391. GUO Zhe, FAN Yang-yu, LIU Shu, LEI Tao, WANG Yi. 3D to 2D: Facial intrinsic shape description maps[J]. Optics and Precision Engineering, 2014, 22(12): 3391.

本文已被 2 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!