高级检索

人脸图像去模糊和三维重建的联合优化方法

Joint Optimization for Deblurring and 3D Reconstruction of Blurry Face Image

  • 摘要: 本文提出了一种对模糊人脸图像的去模糊和三维重建任务进行联合优化的方法.具体为,输入一张模糊人脸图像,交替重复以下步骤: (1) 对人脸图像使用3DMM方法进行三维重建,更新三维人脸模型; (2) 对人脸使用Res-UNet神经网络进行去模糊,并以三维人脸模型再渲染的图像和3DMM系数作为辅助输入,以及原始模糊图像作为矫正输入,得到消除模糊的图像,更新人脸图像.重复上述步骤合适次数后,输出人脸图像和人脸三维模型.这一方法输出的图像清晰度较现有方法获得显著提升,并且能够同时输出良好的三维模型.

     

    Abstract: We propose a joint optimization method for both face deblurring and 3D face reconstruction.With a single blur-ry face image as input, repeat following steps: (1) Apply 3D face reconstruction on the face image, update 3D face model; (2) Deblur the face image by Res-UNet, with re-rendered image and 3DMM coefficients of 3D face model as assist input, and origin blurry face image as correction input, update face image. After repeating above steps for proper number of times, output both deblurred face image and 3D face model. Compared with the ex-isting methods, results of our method get significantly improved.

     

/

返回文章
返回