摘要人脸识别已经越来越成为识别技术的主流之一。基于PCA算法的人脸识别技术也已经成为经典的算法,已经在人脸识别领域被广泛应用。除了基础的PCA算法外,人们根据需要和不同的工作条件,结合了其他优秀的算法对PCA算法进行了改进,从而产生了更多的优秀的基于PCA的人脸识别算法。其中,KPCA算法是其中的一种。67184
本文从PCA和KPCA两种算法的基本原理入手,获得两种算法在人脸识别上的应用步骤。通过比较对于不同样本数、不同样本和相同核函数的不同参数下KPCA算法识别率的变化,我们可以更好地进行样本的选择和核函数参数的选择,最终得到最优的算法选择和最佳的参数选择。
本文所进行的实验主要分为三部分:
1.样本与样本数选择实验:这个实验验证了样本数目对PCA和KPCA两种算法的识别率都不影响,而每组样本的不同会给识别率带来轻微的变化。
2.核函数参数选择实验:这个实验采用的是多项式核函数,函数中包含两个参数,通过实验验证了只有阶次项参数会对KPCA识别率造成较大影响,而且通过识别率比较得出最优的阶次项参数。
3.PCA与KPCA识别率比较实验:这个实验验证了KPCA算法在FERET人脸库上的识别优势比PCA算法要大些。
毕业论文关键词 主元分析法 核主元分析法 核函数 样本数
毕业设计说明书(论文)外文摘要
Title KPCA in the Application of Face Recognition
Abstract
Face recognition technology has become one of the popular methods in the field of recognition increasingly. Face recognition based on principal component analysis(PCA) has become the classic algorithm and is a widespread algorithm on face recognition. In addition to the most basic PCA algorithm, people develop much more excellent face recognition methods based on PCA which are improved by combined with other excellent algorithms, according to the different requirements and working conditions. And Kernel principal component analysis(KPCA) is one of them.
This article begins in the basic principle of PCA and KPCA, and tells us about the procedure of this two methods on face recognition. In order to obtain the most superior solution and the best parameter, we must choose the sample and the parameter of kernel function effectively by comparing the recognition rate’s change in the different quantity of sample, the different sample and the different parameter that is defined in the same KPCA kernel function.
Experiments in this article is pided into 3parts:
1.The experiment about samples and the quantity of samples: this experiment proves that the quantity of samples have no use to the recognition in PCA and KPCA, and different samples cause different recognitions.
2.The experiment about parameter of the kernel function: this experiment uses the polynomial kernel function which has two parameters. After the experiment we prove that only the parameter that decides the order will make an influence on recognition, and find out the best parameter.
3.The experiment about the recognition between PCA and KPCA: this experiment proves that KPCA algorithm is better than PCA algorithm in the FERET face database.
Keywords PCA KPCA kernel function quantity of samples
目 次
1 绪论 1
1.1 本论文的背景和意义 1