基于图像的二维人脸识别技术日趋成熟,但仍受光照、姿态和表情等变化的影响。利用三维人脸模型提高人脸识别性能并将其应用于实际成为近几年学术界的研究趋势。本文提出了SWJTU-MF多模人脸数据库(SWJTU multimodal face database,SWJTU-...基于图像的二维人脸识别技术日趋成熟,但仍受光照、姿态和表情等变化的影响。利用三维人脸模型提高人脸识别性能并将其应用于实际成为近几年学术界的研究趋势。本文提出了SWJTU-MF多模人脸数据库(SWJTU multimodal face database,SWJTU-MF Database),包含200个中性表情中国人的4种人脸样本数据,包括可见光图像、二维视频序列、三维人脸(高精度)和立体视频序列。本文首先分类介绍现有的三维人脸识别算法,然后概述相关的多模人脸数据库,接着提出SWJTU-MF多模人脸数据库,并说明数据库的采集装置、采集环境、采集过程及数据内容,随后简要展示数据标准化过程。最后讨论本数据库面向的应用研究,并给出SWJTU-MF建议的评测协议。展开更多
On the basis of the assumption that the human face belongs to a linear class, a multiple-deformation model is proposed to recover face shape from a few points on a single 2D image. Compared to the conventional methods...On the basis of the assumption that the human face belongs to a linear class, a multiple-deformation model is proposed to recover face shape from a few points on a single 2D image. Compared to the conventional methods, this study has the following advantages. First, the proposed modified 3D sparse deforming model is a noniterative approach that can compute global translation efficiently and accurately. Subsequently, the overfitting problem can be alleviated based on the proposed multiple deformation model. Finally, by keeping the main features, the texture generated is realistic. The comparison results show that this novel method outperforms the existing methods by using ground truth data and that realistic 3D faces can be recovered efficiently from a single photograph.展开更多
文摘基于图像的二维人脸识别技术日趋成熟,但仍受光照、姿态和表情等变化的影响。利用三维人脸模型提高人脸识别性能并将其应用于实际成为近几年学术界的研究趋势。本文提出了SWJTU-MF多模人脸数据库(SWJTU multimodal face database,SWJTU-MF Database),包含200个中性表情中国人的4种人脸样本数据,包括可见光图像、二维视频序列、三维人脸(高精度)和立体视频序列。本文首先分类介绍现有的三维人脸识别算法,然后概述相关的多模人脸数据库,接着提出SWJTU-MF多模人脸数据库,并说明数据库的采集装置、采集环境、采集过程及数据内容,随后简要展示数据标准化过程。最后讨论本数据库面向的应用研究,并给出SWJTU-MF建议的评测协议。
基金the Program for New Century Excellent Talents in University（NCET） The National Natural Science Foundation of China+1 种基金Beijing Natural Science Foundation ProgramBeijing Science and Educational Committee Program.
文摘On the basis of the assumption that the human face belongs to a linear class, a multiple-deformation model is proposed to recover face shape from a few points on a single 2D image. Compared to the conventional methods, this study has the following advantages. First, the proposed modified 3D sparse deforming model is a noniterative approach that can compute global translation efficiently and accurately. Subsequently, the overfitting problem can be alleviated based on the proposed multiple deformation model. Finally, by keeping the main features, the texture generated is realistic. The comparison results show that this novel method outperforms the existing methods by using ground truth data and that realistic 3D faces can be recovered efficiently from a single photograph.
