一个三维人脸真实感模型重建算法  

AN ALGORITHM FOR 3D REALISTIC-LOOKING FACIAL MODEL RECONSTRUCTION

作　　者：梁荣华[1] 陈纯[1] 张慧[1] 

机构地区：[1]浙江大学计算机学院

出　　处：《模式识别与人工智能》2003年第1期116-121,共6页Pattern Recognition and Artificial Intelligence

基　　金：浙江省自然科学基金资助项目(No.601043)

摘　　要：从多幅照片恢复三维人脸模型的难点在于如何自动求取脸部的特征点以及获得逼真的模型。本文提出了从人的头部运动来求取脸部结构的方法,首先使用改进的图像金字塔LK算法匹配对应照片的特征,再由局部向量调整,得到了精确的匹配结果,然后通过相机定标、运动分析、通用模型匹配、纹理映射算法重构出真实感的人脸模型。实验结果及算法分析表明,本文提出的特征匹配算法的精确度较高,实现特定人脸重建方法切实可行,生成的模型可视效果较好。The main issues of 3D facial model reconstruction based on images are in twofold: one is to track facial features automatically and the other is to get the realistic-looking human face model. In this paper, a method on structure from motion is introduced. First the accurate corner matches can be acquired by integrating modified image pyramid Lucas-Kanade algorithm and local adjustment algorithm, then a realistic-looking model is generated by camera calibration, motion analysis and texture mapping. Experimental results demonstrate the high accuracy of the corner matching algirhtm and the vividness of the facial model. Algorithmic analysis shows our approaches can be applied to any human face reconstruction.

关 键 词：三维人脸 真实感模型 特征匹配 运动分析 纹理映射 计算机视觉 人脸重建算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]