人脸图像年龄估计的研究与应用

发布时间：2018-02-22 19:24

本文关键词： 人脸图像年龄估计机器学习特征提取特征融合　出处：《北京工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着计算机技术发展,图像处理技术和人机交互已经成为当今研究的热点之一。以人脸图像为例,人脸图像蕴含着丰富的信息,人的性别、年龄、表情、健康状态等信息都可以从人脸图像中得到。通过人脸图像估计年龄的研究,近年来越来越受到人们的关注。准确地预测人的年龄,从而进行更加深入的研究,改变人们的生活习惯和方式,在信息时代具有重要的意义。本文在FG-NET人脸年龄库的基础上,主要研究了人脸图像年龄估计的特征提取和融合方法。在特征提取阶段,为克服光照产生的影响,本文通过采用均匀局部二值模式(uniform LBP)和Canny边缘检测算子进行图像年龄特征提取。在Canny边缘检测中,进行了阈值优化调整以适应人脸年龄特征的研究。在特征融合阶段,本文采用了直接融合策略和加权融合策略两种算法,通过两种算法的尝试和对比,进行年龄特征融合的研究。在分类学习阶段,本文使用了主成分分析法(PCA)进行了图像数据降维工作;使用了改进后的支持向量机(SVM)进行年龄分类的学习和预测。本文的方法在年龄特征提取上取得了良好的效果。相比于类似研究,本文提出的融合策略较好地表现出了人脸年龄的特征,在预测准确率上取得了一定的提高。在实际应用阶段,本文设计了通过预测人的年龄进行相关信息服务推送的系统。实验证明,本文提出的方法可以较好地克服光照对年龄估计的影响,取得了较为稳定的预测效果。
[Abstract]:With the development of computer technology, image processing technology and human-computer interaction has become a hotspot of current research. Using face image as an example, the face image contains abundant information, the person's gender, age, expression, health status and other information can be obtained from the face image. Through the study of the age estimation of facial images, and in recent years more attention. To accurately predict a person's age, and thus a more in-depth study, changing people's habits and way of life, is of great significance in the information age. Based on the FG-NET facial age library, mainly studies the feature extraction of face image age estimation and fusion method in the feature extraction stage. And in order to overcome the impact of light, through the use of uniform local two value model (uniform LBP) and Canny edge detection operator for image feature extraction. In the age of Canny Edge detection, the threshold optimization in order to adapt to the characteristics of age. In the study of face feature fusion stage, this paper adopts direct fusion method and weighted fusion strategy two algorithms, through trial and comparison of two algorithms, studied the age feature fusion. In the stages of learning classification, this paper uses principal component analysis method (PCA) the image data reduction work; using the improved support vector machine (SVM) learning and prediction of age classification. This method has a good effect in the age feature extraction. Compared with similar studies, the proposed fusion strategy can show the features of face age. The prediction accuracy has improved to some extent. In practical application, this paper designs the system related information services push through the prediction of age. The experimental results show that the proposed method can In order to better overcome the influence of light on age estimation, a more stable prediction effect has been achieved.

【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】