摘要人脸图像的年龄识别是一项很重要的技术,它涉及到许多现实的运用,并且通过人脸图像对年龄进行识别具有很大的挑战。年龄识别会受到诸多干扰因素的影响,如性别、健康、生活方式等等。本文对人脸图片的年龄识别作了基本的介绍,并选用FG-NET数据库的图片作为测试样本,从中挑选了部分作为测试集和训练集。接着对选择的图片作直方图均衡化,直接计算图片的欧氏距离并判别近邻,之后对图片进行人脸提取矫正等预处理后进行测试。后面引入了PCA算法,并运用PCA算法对人脸提取后的图片进行了测试,然后对PCA算法进行了拓展,介绍了2D-PCA,LDA,并且进行了实验。之后介绍了PCA的改进方法——分块PCA,并且运用LBP算法转化图片,对转化前后的图片分别用不同的分块方式进行了测试,同时分析了不同分块方式和LBP算法处理对识别率的影响。27633
关键词 年龄识别 FG-NET数据库 图片预处理 PCA算法 分块PCA
毕业论文设计说明书外文摘要
Title The Human Face Estimation Based on Blocked-Based PCA Algorithm
Abstract
The image-based human age estimation is a significant technology, which involves many real-world applications, and there remains a great challenge through age estimation. Age estimation is influenced by various factors, such as gender, health, lifestyle, etc. This paper introduces image-based human age estimation in basic way, and selects images from FG-NET database as the test sample, part of which are chosen as test list and train list. Then the pretreatment(histogram equalization)is done to all chosen images, and calculate the Euclidean distance to discriminant the nearest neighbor. Then extract the face from all images above and test again. After that, PCA algorithm is introduced to test the extracted images, and PCA is extended to 2D-PCA,LDA,with the experiment done. Then comes the block PCA, which is improved from traditional PCA, and transfer the images by LBP algorithm, the experiments in various block is done to the images transfered, and original images as well, finally analyse the influence of different block and the LBP algorithm.
Keywords age estimation FG-NET database image pretreatment
PCA algorithm block-based PCA
目 次
1 介绍…1
1.1 人脸识别 …1
1.2 人脸年龄识别的发展和存在的问题…1
1.3 人脸年龄自动检测3
2 准备工作和初步实验 4
2.1 FG-NET数据库4
2.2 图片选取…5
2.3 实验工具:matlab5
2.4 人脸图像的年龄识别6
2.5 直方图均衡化7
2.6 人脸提取8
2.7 初步实验…9
3 PCA算法…11
3.1 PCA算法及实验…11
3.2 PCA算法的扩展…16
3.3 评价…19
4 分块PCA…20
4.1 PCA的改进算法——分块PCA20
4.2 LBP算法…22
致谢…24
参考文献…25
1 介绍
人脸是一种非常重要的图像信息,能够高效快捷地传达一系列的信息,而无需通过言辞,使现实生活人们之间的交流更加方便。信息时代的到来以及其高速的发展,也为现代智能系统将人脸识别自动化提供了可能,很明显,人们希望这种识别高效、准确。 基于分块PCA的人脸年龄估计算法:http://www.751com.cn/tongxin/lunwen_22160.html