基于真实感三维头重建的人脸替换

传统的人脸替换技术需要输入的源图像中的人脸与待替换的目标图像中的人脸姿态相似,这限制了人脸替换技术的应用。该文提出一种基于特定三维头模型重建的人脸替换技术。首先,利用替换者的2张正交图像(1张正面图像和1张侧面图像)进行三维头重建。整个三维头重建过程分为人脸图像配准、特征点匹配、标准三维头形变和纹理贴图4个部分。形变部分利用了径向基函数对标准三维头进行形变以获得最终的特定三维头模型。通过建立替换者的真实感三维头模型,渲染出与任意目标图像姿态相一致的人脸图像。为了让渲染后的人脸图像与目标图像进行无缝纹理融合,还进行了色彩传递和基于多分辨率技术的边界融合。用此方法可将故事书中的任意角色替换成用户,以此提供个性化的图书。

Face swapping based on 3-D photo realistic head reconstruction

Traditional face swapping methods require the faces in the input images and the target images to have similar poses and appearance(usually frontal).This limit the application of face swapping.This paper presents a method for face swapping based on personalized 3-D face models.A personalized 3-D face model is used to render a facial image to the same pose as the target image.The 3-D face model is constructed from two user uploaded orthogonal images with one frontal view and one side view.The construction uses image facial alignment,feature point matching,3-D standard head morphing and texture mapping.The final personalized 3-D face is built by morphing a standard 3-D head model using radial basis functions.Color transfer and multi-resolution methods for boundary blurring are used to seamlessly synthesize the face into the image.The technique is used to create a personalized storybook where the characters are replaced with a user’s face.