基于姿态估计的正面人脸图像合成  

Frontal Face Image Synthesis Based on Pose Estimation

作　　者：方三勇 周大可[1] 曹元鹏[1] 杨欣[1] 

机构地区：[1]南京航空航天大学自动化学院,南京210006

出　　处：《计算机工程》2015年第10期240-244,249,共6页Computer Engineering

基　　金：国家自然科学基金资助项目(61172135);南京航空航天大学研究生创新基地(实验室)开放基金资助项目(kfjj20130210)

摘　　要：为对不同姿态下的人脸图像进行处理,提出一种基于姿态估计的正面人脸图像合成方法。利用统计建模的思想重构缺失的人脸形状和纹理。运用平均三维模型估计测试图像的姿态参数,结合压缩感知理论构建形变模型。应用稀疏形变模型理论分别重构测试人脸的三维形状和纹理,根据测试图像与重构模型生成正面人脸图像。实验结果表明,该方法能够由一幅姿态人脸图像合成出精确、自然的正脸图像,并具有较高的识别率。In order to process the face image in different poses,this paper proposes a frontal face image synthesis method based on pose estimation.The method is based on the idea of statistical modeling to reconstruct the missing face shape and texture.Firstly,3D average model is applied to estimate the pose parameters of the test face image.Compressed sensing theory is used to filter prototype samples and then a more accurate model of deformation is built up.Secondly,the test face image is separately expressed by texture vector and shape vector.The deformation model theory is used to reconstruct front texture and shape.Finally,synthesis texture is produced according to the original texture and reconstructed texture.Experimental result shows that this method can be used to synthesize natural frontal face image from non-frontal face image with effectiveness and higher recognition rate.

关 键 词：正面人脸图像合成 平均三维模型 姿态估计 压缩感知 形变模型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]