年龄变化鲁棒的人脸确认算法研究  

Study on the Age-invariant Face Verification

作　　者：牛志玲[1] 王晓燕[1] 潘峰[2] 

机构地区：[1]河南职业技术学院信息工程系,郑州450046 [2]贵州民族大学模式识别与智能系统省级重点实验室,贵阳550025

出　　处：《科学技术与工程》2013年第32期9738-9743,9748,共7页Science Technology and Engineering

基　　金：贵州省科学技术基金(黔科合J字LKM[2011]18号;黔科合计Z字[2009]4002)资助

摘　　要：针对具有较大年龄差的人脸确认系统中的人脸老化问题,提出了一种基于等价局部Gabor二值模式特征的差空间人脸确认算法。首先提取人脸图像的多尺度、多方向的Gabor特征,并对上述特征进行等价LBP运算。其次将LBP图谱进行直方图化,并对直方图进行向量化和差空间运算。最后建立支持向量机(SVM)分类器得出分类结果进而实现身份认证。实验结果表明,在FGnet人脸库上的正确认证率可达94.64%;且对光照与表情变化具有鲁棒性。Considering the problem of face progression in the face verification with lager age differences , a face verification algorithm based on uniform local Gabor binary pattern features in the residual space was presented. Firstly, the multi-scale and multi-orientation Gabor features of face images were extracted and the uniform local bi- nary pattern was operated on each Gabor feature. The histograms were extracted from the LBP features and the his- tograms were converted into vectors to calculate the intra-personal differences and the extra-personal differences. Fi- nally, SVM classifiers were established to acquire the result and complete the identity authentication. The experi- ments demonstrate that the proposed algorithm achieves 94.64% correct verification accuracy on the FGnet. And the above formulation has the strong robustness to the change of illumination and expression.

关 键 词：人脸确认 人脸老化 GABOR 等价局部二值模式 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]