利用关键特征点的多信息融合人脸识别
2017-08-30林克正许松婷李骜吴迪
林克正+许松婷+李骜+吴迪



摘 要:经过多年的人脸识别研究,受光照、噪声等条件影响而导致识别率相对较低的二维人脸识别技术已经跟不上时代最前沿的步伐,虽然三维人脸识别技术正在步步发展,但是它却有着较高的复杂性。针对这一问题,根据传统的深度信息定位方法以及LFA局部特征分析方法,提出了一种改进的三维人脸关键特征点定位算法,并在获取完备聚类训练样本的基础上,进一步提出了全局和局部加权融合的特征提取算法。通过在FRGC和BU3DFE两个人脸库中实验数据的对比分析,本文方法在三维人脸的识别效果方面具有更高的鲁棒性。
关键词:特征点定位;多信息融合;局部特征分析;人脸识别
DOI:1015938/jjhust201703004
中图分类号: TP7511
文献标志码: A
文章编号: 1007-2683(2017)03-0019-06
Abstract:After years of face recognition research, due to the effect of illumination, noise and other conditions have led to the recognition rate is relatively low, 2 d face recognition technology has couldn′t keep up with the pace of The Times the forefront, Although 3 d face recognition technology is developing step by step, but it has a higher complexity In order to solve this problem, based on the traditional depth information positioning method and local characteristic analysis methods LFA, puts forward an improved 3 d face key feature points localization algorithm, and on the basis of the trained sample which obtained by complete cluster, further put forward the global and local feature extraction algorithm of weighted fusion Through FRGC and BU3DFE experiment data comparison and analysis of the two face library, the method in terms of 3 d face recognition effect has a higher robustness
Keywords:locate feature points; multiple information fusion; local characteristic analysis; face recognition
图中我们发现测试对象的识别率整体上比较稳定,对于识别率偏低的对象我们应关注分析其原因。通过表中详细数据我们能够明显看到3号和10号人脸识别率最高,4号和9号人脸识别率较低,而其他测试人脸识别效果处于较为理想状态。
针对两个人脸库中随机选取测试对象出现的识别率较低的情况,我们通过具体数据进行分析[20]。其中一个主要原因就是训练样本图像不完备导致,在对二维虚拟图像进行子集划分时,没有完全的包含不同光照、角度的尽可能多的图像,故导致特征提取时特征向量不够完整,从而导致识别率偏低。另外一种可能是由关键特征点定位错误导致,直接影响了识别效果。……
