摘要: 在运用主成分分析进行人脸识别的过程中,由于实际图像可能符合某种概率密度分布,并且实际用到的图像可能受到不同程度的噪声污染,简单的距离分类已不再适用。基于核函数的最大后验概率分类是将概率密度函数估计中的参数估计、核函数以及贝叶斯理论结合起来,能很好地考虑到概率分布情况,用多元高斯分布下的基于核函数的最大后验概率分类取代距离分类,对于含有不同参数值的高斯噪声图像有较好的识别率。用ORL标准人脸库进行验证,实验结果表明了可行性。
[1] Turk M,Pentland A.Eigenfaces for recognition[J].Journal of Cognitive Neuroscience,1993,3(1):71-86 [2] Tan Ke-ren,Chen Song-can.Adaptively weighted sub-patternPCA for face recognition[J].Neurocomputing,2005,4(3):505-511 [3] Jain A K,Robert P W,et al.Statistical Pattern Recognition:A Review [J].IEEE Transactions on pattern analysis and machine intelligence,2000,2(1):4-37 [4] He Kun,Luan Xin-cheng,et al.Gaussian Noise Removal of Ima-ge on the Local Feature[J].Intelligent Information Technology Application,2008,3(1):867-871 [5] Xu Zeng-lin,Huang Kai-zhu,Zhu Jian-ke,et al.A novel kernel-based maximum a posteriori classification method [J].Neural Networks,2009,2(7):977-987 [6] 张俭鸽,刘洪波.WTPCA和三阶近邻的人脸识别算法仿真[J].计算机工程与应用,2009,45(11):175-177 [7] Motai Y,Yoshida h.Principal Composite Kernel Feature Analysis:Data-Dependent Kernel Approach[J].IEEE Transactions on Knowledge and Date Engineering,2012(99):1-13 [8] 茆诗松,程依明,濮晓龙.概率论与数理统计教程[M].北京:高等教育出版社,2004:287-290 [9] Lukic A S,Wernick M N.Bayesian Kernel Methods for Analysis of Functional Neuroimages[J].IEEE Transactions on Medical Imaging,2007,6(12):1613-1624 |
No related articles found! |
|