一种基于多重影像实现真实感人脸三维重建的方法  

A Novel Way Achieving 3D Reconstruction of Actual Human Face Using Multi Images

作　　者：侯文广[1] 陈大为[2] 丁明跃[1] 

机构地区：[1]华中科技大学图像识别与人工智能研究所,图像信息处理与智能控制教育部重点实验室,湖北武汉430074 [2]香港理工大学工业中心,中国香港

出　　处：《电子学报》2008年第4期661-666,共6页Acta Electronica Sinica

基　　金：中国博士后科学基金(No.20070410282);国家自然科学基金(No.50478058)

摘　　要：基于立体视觉的基本原理,本文提出通过投影不同类型的纹理实现真实人脸几何重建的方法,第一类条纹状的纹理初步建立影像间的视差格网,为第二类精细纹理影像的匹配提供预测值,从而能够快速得到分布均匀且密集可靠的同名点.完成几何重构的基础上,提出基于外接球面实现纹理重建的方法,借助于球面坐标进行纹理影像的选取和颜色处理,减少了重建后纹理颜色的阶跃性变化,达到了匀光的目的.开发了专门的硬件系统,实验表明本文的算法简单,能够取得良好的重建效果.A novel way achieving geometrical reconstruction of actual human face through projecting two types of texture on face in short time is advanced. The first type texture is stripe which is used to establish parallax grid between images. Taking into account of its results, the second type projecting texture is used to match by virtue of its abundant traits. After realizing geometrical reconstruction, the paper provides a general way about achieving actual texture reconstruction by the outer spherical surface sur- rounding object. In order to uniform color, it deals with parts of images in conjunct region and makes the color change meeting a certain function on condition of keeping their original information mostly. Results show this way can improve reconstruction quality and decrease complicacy of algorithm.

关 键 词：人脸重建 三维重建 影像匹配 纹理映射 立体视觉 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]