·378· 智能系统学报 第13卷 表2不同方法在Extended Yale B上的识别准确率 YE Jianfeng,WANG Huaming.An automatic face recogni- Table 2 Recognition rates on the Extended Yale B with tion method using AdaBoost detection and SOM[J].Journal different approaches % of Harbin engineering university,2018,39(1):129-134. 方法名称 S2 S3 S4 S5 AVG [5]SHAN Shiguang,GAO Wen,CAO Bo,et al.Illumination 原图 82.7 19.3 38.0 2.8 23.5 normalization for robust face recognition against varying HEIS lighting conditions[C]//IEEE International Workshop on 83.840.9 12.6 18.1 35.5 Analysis and Modeling of Faces and Gestures.Nice,France: Weber-facel6 100 73.879.1 76.1 81.5 IEEE.2003:157-164. EWGIF+SQI 99.685.384.490.9 90.0 [6]AHONEN T,HADID A,PIETIKAINEN M.Face descrip- 本文方法+SQI 99.191.492.093.093.7 tion with local binary patterns:application to face recogni- tion[J].IEEE transactions on pattern analysis and machine 表3不同方法在Extended Yale B上对一幅图像的平均处 intelligence,2006,28(12y:2037-2041」 理时间 [7]TAN Xiaoyang,TRIGGS B.Enhanced local texture feature Table 3 Average processing time per image on the Exten- ded Yale B with different approaches sets for face recognition under difficult lighting conditions ms [J].IEEE transactions on image processing,2010,19(6): 方法名称 平均处理时间 1635-1650 原图 [8]DENIZ O,BUENO G,SALIDO J,et al.Face recognition HEIS) 0.86 using histograms of oriented gradients[J].Pattern recogni- Weber-face tion letters.2011,32(12)y:1598-1603. 1.95 [9]LUO Yong,GUAN Yepeng.Enhanced facial texture illu- EWGIF+SQ 3.00 mination normalization for face recognition[J.Applied op- 本文方法+SQI 3.01 tics,2015,54(22):6887-6894. [10]GEORGHIADES A S.BELHUMEUR P N.KRIEGMAN 4结束语 D J.From few to many:illumination cone models for face recognition under variable lighting and pose[J].IEEE 本文针对人脸识别中的光照变化问题，提出了 transactions on pattern analysis and machine intelligence, -种基于加权EWGF的人脸光照补偿方法。使用 2001,23(6643-660. 正面光照样本的类间平均脸生成加权系数，作为引 [11]BLANZ V,VETTER T.Face recognition based on fitting a 导滤波损失函数惩罚项的加权系数，配合自商图， 3D morphable model[J].IEEE transactions on pattern ana- 弱化了最终得到的光照补偿图像中人脸平滑区域由 lysis and machine intelligence,2003,25(9):1063-1074. 光照造成的边缘细节噪声。实验结果表明，本文方 [12]ZHANG Lei,SAMARAS D.Face recognition from a single training image under arbitrary unknown lighting us- 法能有效提高人脸识别准确率。未来的工作将深入 ing spherical harmonics[J].IEEE transactions on pattern 研究引导滤波的惩罚项，改进加权系数，优化光照 analysis and machine intelligence,2006,28(3):351-363. 模型，从而进一步提升光照补偿的效果和人脸识别 [13]JOBSON D J,RAHMAN Z,WOODELL G A.Properties 的准确率。 and performance of a center/surround retinex[J].IEEE 参考文献： transactions on image processing,1997,6(3):451-462. [14]ZHANG Taiping,TANG Yuanyan,FANG Bin,et al.Face [1]LIAN Zhichao,ER M J.Illumination normalisation for face recognition under varying illumination using gradientfaces recognition in transformed domain[J].Electronics letters, [J].IEEE transactions on image processing,2009,18(11): 2599-2606. 2010.46(15):1060-1061 [15]WANG Haitao,LI S Z,WANG Yangsheng.Generalized [2]ZOU Jie,JI Qiang,NAGY G.A comparative study of local quotient image[C]//Proceedings of the 2004 IEEE Com- matching approach for face recognition[J].IEEE transac- puter Society Conference on Computer Vision and Pattern tions on image processing,2007,16(10):2617-2628. Recognition.Washington,DC,USA:IEEE,2004,2:II- [3]GUO Guodong,LI SZ,CHAN K.Face recognition by sup- 498-Π-505. port vector machines[Cl//Proceedings of the 4th IEEE Inter- [16]WANG Biao,LI Weifeng,YANG Wenming,et al.Illu- national Conference on Automatic Face and Gesture Recog- mination normalization based on Weber's law with applica- nition.Grenoble.France:IEEE.2000:196-201 tion to face recognition[J].IEEE signal processing letters, [4叶剑锋，王化明.AdaBoost检测结合SOM的自动人脸识 2011,18(8):462-465. 别方法).哈尔滨工程大学学报，2018,39(1)：129-134. [17]BARADARANI A,WU Q M J,AHMADI M.An efficient表 2 不同方法在 Extended Yale B 上的识别准确率 Table 2 Recognition rates on the Extended Yale B with different approaches % 方法名称 S2 S3 S4 S5 AVG 原图 82.7 19.3 38.0 2.8 23.5 HE[5] 83.8 40.9 12.6 18.1 35.5 Weber-face[16] 100 73.8 79.1 76.1 81.5 EWGIF+SQI[19] 99.6 85.3 84.4 90.9 90.0 本文方法+SQI 99.1 91.4 92.0 93.0 93.7 表 3 不同方法在 Extended Yale B 上对一幅图像的平均处 理时间 Table 3 Average processing time per image on the Exten￾ded Yale B with different approaches ms 方法名称 平均处理时间 原图 — HE[5] 0.86 Weber-face[16] 1.95 EWGIF+SQI[19] 3.00 本文方法+SQI 3.01 4 结束语 本文针对人脸识别中的光照变化问题，提出了 一种基于加权 EWGIF 的人脸光照补偿方法。使用 正面光照样本的类间平均脸生成加权系数，作为引 导滤波损失函数惩罚项的加权系数，配合自商图， 弱化了最终得到的光照补偿图像中人脸平滑区域由 光照造成的边缘细节噪声。实验结果表明，本文方 法能有效提高人脸识别准确率。未来的工作将深入 研究引导滤波的惩罚项，改进加权系数，优化光照 模型，从而进一步提升光照补偿的效果和人脸识别 的准确率。 参考文献： LIAN Zhichao, ER M J. Illumination normalisation for face recognition in transformed domain[J]. Electronics letters, 2010, 46(15): 1060–1061. [1] ZOU Jie, JI Qiang, NAGY G. A comparative study of local matching approach for face recognition[J]. IEEE transac￾tions on image processing, 2007, 16(10): 2617–2628. [2] GUO Guodong, LI S Z, CHAN K. Face recognition by sup￾port vector machines[C]//Proceedings of the 4th IEEE Inter￾national Conference on Automatic Face and Gesture Recog￾nition. Grenoble, France: IEEE, 2000: 196–201. [3] 叶剑锋, 王化明. AdaBoost 检测结合 SOM 的自动人脸识 别方法[J]. 哈尔滨工程大学学报, 2018, 39(1): 129–134. [4] YE Jianfeng, WANG Huaming. An automatic face recogni￾tion method using AdaBoost detection and SOM[J]. Journal of Harbin engineering university, 2018, 39(1): 129–134. SHAN Shiguang, GAO Wen, CAO Bo, et al. Illumination normalization for robust face recognition against varying lighting conditions[C]//IEEE International Workshop on Analysis and Modeling of Faces and Gestures. Nice, France: IEEE, 2003: 157–164. [5] AHONEN T, HADID A, PIETIKAINEN M. Face descrip￾tion with local binary patterns: application to face recogni￾tion[J]. IEEE transactions on pattern analysis and machine intelligence, 2006, 28(12): 2037–2041. [6] TAN Xiaoyang, TRIGGS B. Enhanced local texture feature sets for face recognition under difficult lighting conditions [J]. IEEE transactions on image processing, 2010, 19(6): 1635–1650. [7] DÉNIZ O, BUENO G, SALIDO J, et al. Face recognition using histograms of oriented gradients[J]. Pattern recogni￾tion letters, 2011, 32(12): 1598–1603. [8] LUO Yong, GUAN Yepeng. Enhanced facial texture illu￾mination normalization for face recognition[J]. Applied op￾tics, 2015, 54(22): 6887–6894. [9] GEORGHIADES A S, BELHUMEUR P N, KRIEGMAN D J. From few to many: illumination cone models for face recognition under variable lighting and pose[J]. IEEE transactions on pattern analysis and machine intelligence, 2001, 23(6): 643–660. [10] BLANZ V, VETTER T. Face recognition based on fitting a 3D morphable model[J]. IEEE transactions on pattern ana￾lysis and machine intelligence, 2003, 25(9): 1063–1074. [11] ZHANG Lei, SAMARAS D. Face recognition from a single training image under arbitrary unknown lighting us￾ing spherical harmonics[J]. IEEE transactions on pattern analysis and machine intelligence, 2006, 28(3): 351–363. [12] JOBSON D J, RAHMAN Z, WOODELL G A. Properties and performance of a center/surround retinex[J]. IEEE transactions on image processing, 1997, 6(3): 451–462. [13] ZHANG Taiping, TANG Yuanyan, FANG Bin, et al. Face recognition under varying illumination using gradientfaces [J]. IEEE transactions on image processing, 2009, 18(11): 2599–2606. [14] WANG Haitao, LI S Z, WANG Yangsheng. Generalized quotient image[C]//Proceedings of the 2004 IEEE Com￾puter Society Conference on Computer Vision and Pattern Recognition. Washington, DC, USA: IEEE, 2004, 2: II- 498–II-505. [15] WANG Biao, LI Weifeng, YANG Wenming, et al. Illu￾mination normalization based on Weber's law with applica￾tion to face recognition[J]. IEEE signal processing letters, 2011, 18(8): 462–465. [16] [17] BARADARANI A, WU Q M J, AHMADI M. An efficient ·378· 智 能 系 统 学 报 第 13 卷