基于改进PCA算法的人脸识别
2018-02-01张杨张仁杰
张杨+张仁杰



摘要:由于经典的PCA算法要求样本满足高斯分布,然而现实中的样本往往因为表情、角度、光照等原因不满足高斯分布,导致算法识别率不高。因此,提出一种基于改进PCA算法的人脸识别方法。首先,将具有相似特征(表情、角度、亮度)的不同样本通过分块方式划分在一个矩阵中,使样本趋于高斯分布;其次,通过直方图均衡化样本的方法,加强样本对比度,以突出样本的人脸器官特征;最后采用经典PCA算法进行辨识。通过在ORL人脸库上的实验得出,该方法不但耗费总时间少于经典的PCA算法,而且识别率也得到提升,具有一定可行性。
关键词:人脸识别;分块;直方图均衡化;改进的PCA算法
DOIDOI:10.11907/rjdk.172191
中图分类号:TP301
文献标识码:A文章编号文章编号:16727800(2018)001003203
Abstract:The classical PCA algorithm requires the sample to satisfy the Gaussian distribution, but the real samples often do not satisfy the Gaussian distribution because of the expression, the angle and the light. So the recognition rate of this algorithm is not high. For this reason, this paper presents a face recognition method based on improved PCA algorithm. Firstly, different samples with similar characteristics (expression, angle, brightness) are divided into a matrix by way of block in order to make samples tend to Gaussian distribution. Secondly, through the method of histogram to equalize the sample, the contrast of the sample is enhanced to highlight the facial features. Finally, the classical PCA algorithm is used to identify the samples. And through the experiment on the ORL face database, this method not only cost less total time than the classic PCA algorithm and recognition rate has also been improved. In a general, this way is feasible.
Key Words:face recognition; block; histogram equalization; improved PCA algorithm
0引言
随着现代信息化技术的迅速发展,人脸识别技术也进入快速发展期。人脸识别技术目前已被应用于各个行业,包括公安的刑侦破案、网络信息安全、机器人智能化等领域。不仅如此,人脸识别技术具有其它方法所不具备的优势,如非接触式、自然性。因此,用户的可接受度高,具有广阔的发展前景[13]。
经典PCA(Principle Component Analysis)算法是人脸识别中的常用算法,此方法通过KL变换提取人脸特征构成特征脸空间,在识别时将待识别的样本投影到此特征脸空间,可得到一组投影向量,再通过与数据库中的每个样本比较进行识别。经典的PCA算法受人脸表情、角度、光照等因素影响较大[4],为了减少影响,本文改进了经典的PCA算法。本文算法基于被广泛使用的线性鉴别方法——PCA算法[5],一方面,通过分块方式将具有相似特征的样本划分在同一矩陣,再直方图均衡化样本,找到总体最小区域后,采用经典PCA算法对预处理过的样本进行特征提取及识别。……
