第13卷第3期 智能系统学报 Vol.13 No.3 2018年6月 CAAI Transactions on Intelligent Systems Jun.2018 D0:10.11992/tis.201710014 网络出版地址：http:/kns.cnki.net/kcms/detail/23.1538.TP.20180404.0942.002.html 一种新融合算法的维吾尔族人脸识别 伊力哈木·亚尔买买提 (新疆大学电气工程学院，新疆乌鲁木齐830047) 摘要：针对维吾尔族人脸在光照以及部分遮挡下的辨识率下降和鲁棒性差的问题，提出了二维离散余弦变换 (2DDCT)与方向边缘幅值模式(POEM)相融合的维吾尔族人脸识别算法。首先，把维吾尔族人脸图像分块处理，并 使用2DDCT把其分块后的维吾尔族人脸图像转换为频域状态；其次，压缩维吾尔族人脸图像以排除维吾尔族人脸图 像中无用信息，即中频部分与非低频部分，并进行二维离散余弦逆变换(DCT)得到重构的维吾尔族人脸图像；然后， 经POEM计算维吾尔族人脸图像的特征量得到其相应的POEM直方图并把直方图级联在一起，作为该中心特征点 的POEM纹理直方图，得到维吾尔族人脸特征点的纹理特征信息：最后，采用深度学习算法进行分类识别。本文通 过实验提出的算法，在自建的维吾尔族人脸库中能够进一步提高其人脸识别率，在维吾尔族人脸数据库中其运算速 度也有很大提高。实验结果表明，该算法尤其是在维吾尔族人脸数据库中拥有较好的识别精度，具有很强的鲁棒性， 特别是在光照以及部分遮挡下具有很强的优势。 关键词：人脸识别：维吾尔族；光照；遮挡；离散余弦变换：方向边缘幅值模式；频域状态；深度学习 中图分类号：TP391.4文献标志码：A文章编号：1673-4785(2018)03-0431-06 中文引用格式：伊力哈木·亚尔买买提.一种新融合算法的维吾尔族人脸识别.智能系统学报，2018,13(3)：431-436. 英文引用格式：Yilihamu-Yaermaimaiti.A new fusion algorithm for uyghur face recognition[J.CAAI transactions on intelligent systems,2018.13(3:431-436. A new fusion algorithm for uyghur face recognition Yilihamu Yaermaimaiti (College of Electncian Engineering,Xinjiang University,Urumqi 830047,China) Abstract:Considering the inferior robustness of Uyghur face recognition under illumination and partial occlusion,this study proposes a Uyghur face recognition algorithm based on two-dimensional discrete cosine transform(2DDCT)and patterns of oriented edge magnitudes(POEM).The Uygur face images were partitioned into several blocks,and 2DDCT was used to transform the partitioned images into a frequency domain.The images were compacted and irrelevant in- formation was excluded,i.e.,the medium-frequency portion and the low-frequency portion,and then a two-dimensional inverse discrete cosine transform(IDCT)was carried out to obtain a reconstructed Uygur face image.The POEM was then used to calculate the characteristic quantity of the Uygur face image to obtain the corresponding POEM histogram. All histograms were cascaded together as the POEM texture histogram of the central characteristic point to acquire the texture feature information of Uygur face feature point.Finally,a deep learning algorithm was used to classify recogni- tion.The algorithm proposed in this paper can improve the face recognition rate and operation speed of a self-built Uyghur face database.Experimental results show that the algorithm has good recognition accuracy,especially for a Uyghur face database,and strong robustness,especially under illumination and partial occlusion. Keywords:face recognition;uyghur,illumination;occlusion;dct,poem;frequency domain state;deep learning 收稿日期：2017-10-23.网络出版日期：2018-04-04 新疆位于中国的西北部地区，具有独特的地理 基金项目：国家自然科学基金项目(61462082). 通信作者：伊力哈木-亚尔买买提.E-mail:65891080@q9.com. 位置，少数民族众多，其中维吾尔族属于新疆最大DOI: 10.11992/tis.201710014 网络出版地址: http://kns.cnki.net/kcms/detail/23.1538.TP.20180404.0942.002.html 一种新融合算法的维吾尔族人脸识别 伊力哈木·亚尔买买提 （新疆大学 电气工程学院，新疆 乌鲁木齐 830047） 摘 要：针对维吾尔族人脸在光照以及部分遮挡下的辨识率下降和鲁棒性差的问题，提出了二维离散余弦变换 (2DDCT) 与方向边缘幅值模式 (POEM) 相融合的维吾尔族人脸识别算法。首先，把维吾尔族人脸图像分块处理，并 使用 2DDCT 把其分块后的维吾尔族人脸图像转换为频域状态；其次，压缩维吾尔族人脸图像以排除维吾尔族人脸图 像中无用信息，即中频部分与非低频部分，并进行二维离散余弦逆变换 (IDCT) 得到重构的维吾尔族人脸图像；然后， 经 POEM 计算维吾尔族人脸图像的特征量得到其相应的 POEM 直方图并把直方图级联在一起，作为该中心特征点 的 POEM 纹理直方图，得到维吾尔族人脸特征点的纹理特征信息；最后，采用深度学习算法进行分类识别。本文通 过实验提出的算法，在自建的维吾尔族人脸库中能够进一步提高其人脸识别率，在维吾尔族人脸数据库中其运算速 度也有很大提高。实验结果表明，该算法尤其是在维吾尔族人脸数据库中拥有较好的识别精度，具有很强的鲁棒性， 特别是在光照以及部分遮挡下具有很强的优势。 关键词：人脸识别；维吾尔族；光照；遮挡；离散余弦变换；方向边缘幅值模式；频域状态；深度学习 中图分类号：TP391.4 文献标志码：A 文章编号：1673−4785(2018)03−0431−06 中文引用格式：伊力哈木·亚尔买买提. 一种新融合算法的维吾尔族人脸识别[J]. 智能系统学报, 2018, 13(3): 431–436. 英文引用格式：Yilihamu·Yaermaimaiti . A new fusion algorithm for uyghur face recognition[J]. CAAI transactions on intelligent systems, 2018, 13(3): 431–436. A new fusion algorithm for uyghur face recognition Yilihamu·Yaermaimaiti (College of Electncian Engineering, Xinjiang University, Urumqi 830047, China) Abstract: Considering the inferior robustness of Uyghur face recognition under illumination and partial occlusion, this study proposes a Uyghur face recognition algorithm based on two-dimensional discrete cosine transform (2DDCT) and patterns of oriented edge magnitudes (POEM). The Uygur face images were partitioned into several blocks, and 2DDCT was used to transform the partitioned images into a frequency domain. The images were compacted and irrelevant in￾formation was excluded, i.e., the medium-frequency portion and the low-frequency portion, and then a two-dimensional inverse discrete cosine transform (IDCT) was carried out to obtain a reconstructed Uygur face image. The POEM was then used to calculate the characteristic quantity of the Uygur face image to obtain the corresponding POEM histogram. All histograms were cascaded together as the POEM texture histogram of the central characteristic point to acquire the texture feature information of Uygur face feature point. Finally, a deep learning algorithm was used to classify recogni￾tion. The algorithm proposed in this paper can improve the face recognition rate and operation speed of a self-built Uyghur face database. Experimental results show that the algorithm has good recognition accuracy, especially for a Uyghur face database, and strong robustness, especially under illumination and partial occlusion. Keywords: face recognition; uyghur; illumination; occlusion; dct; poem; frequency domain state; deep learning 新疆位于中国的西北部地区，具有独特的地理 位置，少数民族众多，其中维吾尔族属于新疆最大 收稿日期：2017−10−23. 网络出版日期：2018−04−04. 基金项目：国家自然科学基金项目 (61462082). 通信作者：伊力哈木·亚尔买买提. E-mail：65891080@qq.com. 第 13 卷第 3 期 智 能 系 统 学 报 Vol.13 No.3 2018 年 6 月 CAAI Transactions on Intelligent Systems Jun. 2018