高级检索

从多张非标定图像重建三维人脸

Reconstructing 3D Face from Multiple Uncalibrated Images

  • 摘要: 为了在人脸动画中方便地重建真实感强的三维人脸模型,提出一种从多角度、非标定图像重建三维人脸的方法.首先利用一种基于规则网格变形的模型归一化方法建立人脸形变模型;然后分别在5幅人脸图像上手动标注不超过14个关键点,通过拟合图像上的关键点重建个性化的人脸几何形状;最后从每个角度渲染出一幅形状无关纹理,并将5幅形状无关纹理融合成一幅平滑的纹理图像.重建结果显示,该方法能在较少的手动交互的情况下重建出具有真实感的个性化人脸模型.

     

    Abstract: In order to reconstruct realistic 3D face model in facial animation,this paper presents a method to reconstruct 3D face model from multiple uncalibrated images.First,we build morphable models through a model normalization method based on regular grid deformation.Second,we label no more than 14 key points manually on each image,and reconstruct individual face geometry by fitting the key points on the face images.Finally,we render the face model into shape-free texture on each view,and fuse the five shape-free textures into one smooth texture image.The reconstruction results show that our method can generate realistic individual face model in the case of less manual interaction.

     

/

返回文章
返回