摘要
为了实现基于单张2维(2D)正面人脸图像的3维(3D)人脸重构,采用基于形状和纹理映射的3D人脸重构方法。首先通过主动形状模型及显式形状回归等算法提取预处理后2D人脸图像的特征点及其坐标和纹理信息,然后在标准3D人脸模型Candide-3上建立2D人脸图像与3D人脸模型之间稠密对应的映射关系,再利用纹理映射方法对适配后的3D模型进行覆盖,以便得到与真实人脸高度相似的3D人脸模型,最后使用Open GL将2D图像作为纹理映射至目标3D人脸模型上成为一个真实的3D人脸。实验结果表明:该方法得到的3D人脸模型仍然保持着与原始通用人脸模型相似的形状,并具有良好的连续性和光滑性。
In order to realize 3 dimensional(3D) face reconstruction based on 2 dimensional(2D)frontal face image,this paper adopts a 3D face reconstruction method of shape and texture mapping based on the 2D frontal face image. Firstly,face image preprocessing,feature point extraction,2D face image pretreatment,and the coordinates and texture information are all obtained by through the active shape model and the explicit shape regression algorithm,that is,there are 31 performancefeature information of the feature points are marked in the eyes,eyebrows,nose,mouth and other parts of the 2D frontal face image. Then,the standard 3D face model Candide-3 will be used to set up the mapping relationship between the dense 2D face image and the 3D face model by corresponding the 31 feature points accurately mapped on the 3D model,so that the 3D model with only statistical characteristics and the standard 3D face model without non personal characteristics becomes rich individuality and authenticity. Then,the face information extracted from the adaptation process,and the 3D model fits the cover with the texture mapping method,and the 3D face model is highly similar to the real human face obtained. We used texture mapping method to cover the 3D model after adaptation,so as to obtain 3D face model with the real face height. Finally,Open GL is used to the2D image as texture mapping to the target 3D face model,and the 3D face model is deformed into a real 3D face. The experimental results show that the 3D face model interpolation method used in this paper remains similar with the original shape of the generic face model, and which has good smoothness and continuity.
作者
高昊烨
汤艺璇
陈江源
叶超
肖南峰
GAO Haoye;TANG Yixuan;CHEN Jiangyuan;YE Chao;XIAO Nanfeng(School of Computer Science and Engineering,South China University of Technology, Guangzhou 510006, China)
出处
《重庆理工大学学报(自然科学)》
CAS
北大核心
2018年第4期161-170,共10页
Journal of Chongqing University of Technology:Natural Science
基金
国家自然科学基金资助项目(61573145)
广东省公益研究与能力建设专项资金项目(2014B010104001)
广东省自然科学基金项目(2015A030308018)
关键词
3D人脸重建
主动形状模型
特征提取
径向基函数
纹理映射
3D face reconstruction
active shape model
feature extraction
radial basis function
texture mapping