素描梯度方向直方图用于人脸画像识别  

Sketch-based histogram of orientation gradient for face sketch recognition

作　　者：李伟红[1] 付伟锋 张镇[1] 龚卫国[1] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室重庆大学光电工程学院,重庆400044

出　　处：《仪器仪表学报》2015年第2期368-376,共9页Chinese Journal of Scientific Instrument

基　　金：国家科技惠民计划(2013GS500303);重庆市重点科技计划(CSTC2012GG-YYJSB40001;CSTC2013-JCSF40009)资助项目

摘　　要：为有效提取人脸画像的轮廓信息,首先提出素描算子,使人脸画像的梯度幅值图像清晰、厚重,由此形成素描梯度方向直方图(S-HOG)描述子。其次优化S-HOG描述子的主要参数,提升S-HOG描述子的图像信息表达能力。针对缺乏实验用人脸刑侦画像库问题,建立了由不同画家完成的多风格人脸画像(MSFS)库,用于模拟人脸刑侦画像库。最后,在MSFS库、人脸观察画像库、PRIP人脸合成画像库上与尺度不变特征转换(SIFT)描述子、统一局部二值模式(ULBP)描述子及HOG描述子进行了比较实验。结果表明:在人脸观察画像库上,提出的S-HOG描述子比SIFT、HOG和ULBP描述子的识别率提高5.14%、3.22%和68.81%。在MSFS库上,分别提高14%、5.5%和26.5%。在PRIP库上,分别提高6.15%、4.06%和9.75%。In order to extract the outline information of face sketch effectively,a sketch operator that clearly extracts face sketch＇s thick gradient amplitude image contour is proposed,which develops sketch histogram of orientation gradient（ S-HOG） descriptor.The main parameters in the process of S-HOG descriptor accumulating image gradient information are optimized,which improves image information description ability of S-HOG descriptor.Aiming at addressing the problem of lacking of public forensic sketch database in face sketch recognition research,a multiple styles face sketch（ MSFS） database completed by different artists is set up,used for simulates forensic sketch database.MSFS database,viewed face sketch database,composite sketch database created by pattern recognition and image processing（ PRIP） lab are used in experiments.With scale invariant feature transform（ SIFT）,Uniform local binary pattern（ ULBP） and HOG are performed the compared experiments.The results show,in the viewed face sketch database,the recognition rate of S-HOG is improved by 5.14%,3.22% and 68.81% than SIFT,HOG and ULBP descriptors＇.In the MSFS database,it is improved by 14%,5.5% and 26.5% respectively.In the PRIP composite sketch database,it also is improved by 6.15%,4.06% and 9.75%.

关 键 词：人脸画像 素描算子 梯度方向直方图描述子 不变特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]