摘要
针对目前三维人脸几何特征识别算法中计算量大和设备昂贵,尤其是在特征融合时加权值确定的不精确性问题,提出了根据双目立体视觉原理,通过对普通二维图像确定脸部关键部位特征点的三维几何特征信息,并且依照类内距离越小越好,类间距离越大越好的准则设定适应度函数,使用人脸样本数据根据遗传算法进行训练,得到使适应度函数最小时的最优解,从而获得三维人脸几何特征融合时的最佳加权值。实验结果表明了该算法的可行性和有效性。
Expensive instruments and great computation, especially the imprecision of fusing and weighting geometric features, all the those problems will happen when using the recognition algorithm of 3D geometric features. In order to address the issue, a method is presented, which only the pivotal facial geometric features are used by the theory of binocular stereo vision and the common 2D pictures, and the fitness function is set by the rule that the within-class distance as small as possible and the class distance as large as possible , and then the sample data of faces are used to train by the genetic algorithms to get the optimal solution of which the fitness function is minmum, so the best weight of fusing geometric features is obtained. The feasibility and ef fectiveness of the method are proved by the experiment.
出处
《计算机工程与设计》
CSCD
北大核心
2012年第11期4328-4332,共5页
Computer Engineering and Design
基金
广西自然科学基金项目(2011GXNSFA018158)
广西科技开发基金项目(桂科攻11107006-45)
关键词
三维人脸识别
几何特征
遗传算法
类内距离
类间距离
双目立体视觉
3D face recognition
geometric features
genetic algorithms
within-classs distance
class distance
binocularstereo vision
