基于多尺度分析的人脸识别算法研究  

Research on face recognition algorithm based on multiscale analysis

作　　者：赵文仓[1] 王倩[1] 董瑞霞[1] 

机构地区：[1]青岛科技大学自动化与电子工程学院,山东青岛266042

出　　处：《工业仪表与自动化装置》2017年第4期11-15,共5页Industrial Instrumentation & Automation

基　　金：山东省科技计划项目(2013YD01033);山东省自然科学基金(ZR2015FL008)

摘　　要：针对传统的人脸识别算法易受光照等因素影响的缺点,提出了一种基于多尺度分析的人脸识别算法。首先,对采集到的人脸图像进行预处理,去除噪声,减弱无用信息,增强有用信息;然后,在粗尺度上采用形态学梯度边缘检测算法对人脸轮廓进行提取,缩小人脸库的搜索范围;最后,在细尺度上对人脸的不变特征进行提取,采用Harris角点检测算法对在粗尺度上得到的边缘图像进行特征提取,减少计算量,提高了识别速度。在所拍摄的人脸库上对算法进行验证,实验表明,对易于进行轮廓提取的人脸图像的识别速度较快,精度较高。The traditional face recognition algorithms are easily affected by some factors such as illumination,so the face recognition algorithm based on multiscale analysis is proposed in this paper.First,preprocessing on the detected face images is conducted.It can remove noise,reduce the useless information and enhance the useful information.Then,the face contour feature is extracted through the morphology gradient edge detection algorithm on the coarse scale.It can narrow down the search scope in the face database.Finally,the invariant feature is extracted on the fine scale.The feature of the edge images adopted on the coarse scale is extracted through the Harris corner detection algorithm.The calculation is reduced and the speed of recognition is improved.This algorithm is verified on the face database photographed by myself.Experimental results show that the algorithm has faster recognition speed and higher precision on the face images whose contour feature is easily extracted.

关 键 词：人脸识别 多尺度分析 轮廓特征 角点特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]