摘要
针对传统人脸检测方法采用空间向量对复杂环境下的高维度人脸特征进行辨识时,存在检测效率低、检测精度差的问题,提出一种基于随机Fourier有监督特征变换降维算法的人脸检测方法.首先,通过随机Fourier映射随机形成大规模多维候选集合,采用特征选择算法获取特征集内的最佳子集;其次,基于l_(2,1)范数的极限学习机,产生高斯核拟合效果的随机映射,利用l_(2,1)正规则化过滤掉人脸随机特征中的无价值及冗余特征,并对该过程进行优化,提高人脸特征降维的精度;最后,采用基于降维特征与Adaboost算法的人脸检测方法获取的降维特征,通过Boosted级联算法获取级联分类器,实现人脸特征的准确检测.实验结果表明,该方法的漏检率和误检率均为8%,平均检测时间为118ms,运行效率和检测精度均较高.
Aiming at the problem that traditional face detection methods used space vectors to identify high-dimensional face features in complex environments,which had low detection efficiency and poor detection accuracy,the author proposed a face detection method based on dimensionality reduction algorithm of random Fourier supervised feature transformation.Firstly,large-scale multi-dimensional candidate sets were randomly formed by random Fourier mapping,and the best subset of feature sets was obtained by feature selection algorithm.Secondly,based on the extreme learning machine of l_(2,1) norm,the random mapping of Gauss kernel fitting effect was generated,the worthless and redundant features of face random features were filtered by l_(2,1) normalization,and the process was optimized to improve the accuracy of face feature dimensionality reduction.Finally,the dimensionality reduction feature based on dimensionality reduction feature and Adaboost's face detection algorithm was adopted,and the cascade classifier was obtained by Boosted cascade algorithm to realize the accurate detection of face features.The experimental results show that the missing detection rate and false detection rate of the proposed method are both 8%,and the average detection time is 118 ms.The proposed method has high operational efficiency and detection accuracy.
引文
[1]刘帅,林克正,孙旭东,等.基于聚类的SIFT人脸检测算法[J].哈尔滨理工大学学报,2014,19(1):31-35.(LIU Shuai,LIN Kezheng,SUN Xudong,et al.Scale-Invariant Feature Transform Based on Clustering in Face Recognition[J].Journal of Harbin University of Science and Technology,2014,19(1):31-35.)
[2]姚明海,王娜,易玉根,等.基于有监督降维的人脸识别方法[J].计算机工程,2014,40(5):228-233.(YAO Minghai,WANG Na,YI Yugen,et al.Face Recognition Method Based on Supervised Dimensionality Reduction[J].Computer Engineering,2014,40(5):228-233.)
[3]柴瑞敏,曹振基.基于Gabor小波与深度信念网络的人脸识别方法[J].计算机应用,2014,34(9):2590-2594.(CHAI Ruimin,CAO Zhenji.Face Recognition Algorithm Based on Gabor Wavelet and Deep Belief Networks[J].Journal of Computer Applications,2014,34(9):2590-2594.)
[4]翟社平,李炀,马蒙雨,等.基于LBP和SVM的人脸检测[J].计算机技术与发展,2017,27(9):44-47.(ZHAI Sheping,LI Yang,MA Mengyu,et al.Face Detection Based on LBP and SVM[J].Computer Technology and Development,2017,27(9):44-47.)
[5]李燕,王玲.基于肤色和Haar方差特征的人脸检测[J].计算机工程与科学,2015,37(1):146-151.(LI Yan,WANG Ling.Face Detection Algorithm Based on Skin Color and Haar Variance Characteristics[J].Computer Engineering and Science,2015,37(1):146-151.)
[6]张小飞,张立岑,陈未央,等.MIMO阵列中基于PM和降维变换的高效DOA估计算法[J].数据采集与处理,2014,29(3):372-377.(ZHANG Xiaofei,ZHANG Licen,CHEN Weiyang,et al.Computationally Efficient DOAEstimation for MIMO Array Using Propagator Method and Reduced-Dimension Transformation[J].Journal of Data Acquisition&Processing,2014,29(3):372-377.)
[7]董九玲,赖惠成,杨敏,等.基于图像旋转变换的改进PCA与LVQ的人脸识别[J].激光杂志,2015,36(9):51-55.(DONG Jiuling,LAI Huicheng,YANG Min,et al.Face Recognition Method with Improved PCA and LVQ Network Based on Image Rotation[J].Laser Journal,2015,36(9):51-55.)
[8]王庆伟,应自炉.一种基于Haar-Like T特征的人脸检测算法[J].模式识别与人工智能,2015,28(1):35-41.(WANG Qingwei,YING Zilu.A Face Detection Algorithm Based on Haar-Like T Features[J].Pattern Recognition and Artificial Intelligence,2015,28(1):35-41.)
[9]缪丹权,郑河荣,顾国民.基于优化加权参数的AdaBoost人脸检测算法[J].计算机工程与应用,2014,50(19):173-177.(MIU Danquan,ZHENG Herong,GU Guomin.AdaBoost Face Detection Algorithm Based on Optimized Weighting Parameter[J].Computer Engineering and Applications,2014,50(19):173-177.)
[10]王小玉,张亚洲,陈德运.基于多块局部二值模式特征和人眼定位的人脸检测[J].仪器仪表学报,2014(12):2739-2745.(WANG Xiaoyu,ZHANG Yazhou,CHEN Deyun.Face Detection Based on MB-LBP and Eye Tracking[J].Chinese Journal of Scientific Instrument,2014(12):2739-2745.)