智能系统学报
智能繫統學報
지능계통학보
CAAI TRANSACTIONS ON INTELLIGENT SYSTEMS
2015年
1期
12-19
,共8页
阮晓虎%李卫军%覃鸿%董肖莉%张丽萍
阮曉虎%李衛軍%覃鴻%董肖莉%張麗萍
원효호%리위군%담홍%동초리%장려평
人脸识别%图像规格化%配准判断%图像特征%SIFT描述子%梯度方向直方图%关键点定位%图像匹配
人臉識彆%圖像規格化%配準判斷%圖像特徵%SIFT描述子%梯度方嚮直方圖%關鍵點定位%圖像匹配
인검식별%도상규격화%배준판단%도상특정%SIFT묘술자%제도방향직방도%관건점정위%도상필배
face recognition%image normalization%alignment assessment%image feature%SIFT descriptor%gradient orientation histogram%key point location%image matching
现有的人脸识别应用系统大都忽略了人脸配准的检查,造成“误配准灾难”,导致识别性能下降。因此,对规格化后的人脸图像进行判断筛选,以保证只有正确配准的人脸图像才能用于后续识别。选用一定数量正确配准的规格化人脸图像平均值作为标准人脸,用SIFT关键点定位方法得到标准人脸的多个关键点,采用分块的梯度方向直方图统计方法提取关键点的邻域图像特征;然后,将标准人脸的关键点位置作为待检测人脸的定位点,用同样的方法提取定位点的邻域图像特征;计算待检图像与标准人脸图像对应关键点的特征矢量相似度,设定合理阈值判断待检测图像是否配准。实验证明,该方法能有效去除误配准人脸图像,有利于提高人脸识别系统的可靠性。
現有的人臉識彆應用繫統大都忽略瞭人臉配準的檢查,造成“誤配準災難”,導緻識彆性能下降。因此,對規格化後的人臉圖像進行判斷篩選,以保證隻有正確配準的人臉圖像纔能用于後續識彆。選用一定數量正確配準的規格化人臉圖像平均值作為標準人臉,用SIFT關鍵點定位方法得到標準人臉的多箇關鍵點,採用分塊的梯度方嚮直方圖統計方法提取關鍵點的鄰域圖像特徵;然後,將標準人臉的關鍵點位置作為待檢測人臉的定位點,用同樣的方法提取定位點的鄰域圖像特徵;計算待檢圖像與標準人臉圖像對應關鍵點的特徵矢量相似度,設定閤理閾值判斷待檢測圖像是否配準。實驗證明,該方法能有效去除誤配準人臉圖像,有利于提高人臉識彆繫統的可靠性。
현유적인검식별응용계통대도홀략료인검배준적검사,조성“오배준재난”,도치식별성능하강。인차,대규격화후적인검도상진행판단사선,이보증지유정학배준적인검도상재능용우후속식별。선용일정수량정학배준적규격화인검도상평균치작위표준인검,용SIFT관건점정위방법득도표준인검적다개관건점,채용분괴적제도방향직방도통계방법제취관건점적린역도상특정;연후,장표준인검적관건점위치작위대검측인검적정위점,용동양적방법제취정위점적린역도상특정;계산대검도상여표준인검도상대응관건점적특정시량상사도,설정합리역치판단대검측도상시부배준。실험증명,해방법능유효거제오배준인검도상,유리우제고인검식별계통적가고성。
The lacking of confirmation for face alignment leads to an incorrect feature match .The decline of recogni-tion rate in current application of face recognition is called "mis-alignment crash".Therefore , it is necessary to test and filter the normalized face images to make sure only the aligned face images can go through the recognition pro -cedure .In the method , a bunch of right-alignment normalized face images were used to form a mean face which was defined as the standard face .The key points location theory of SIFT was used to get the key points of standard face and the features of neighboring images were extracted on the basis of blocked statistical histogram in gradient orien -tation .The location of key points of a standard face was taken as the positioning point of a face to be detected .Using the same method to extract the features of neighboring images showed that the similarities of the test images to the standard face were calculated according to their corresponding feature descriptors of the key points .A reasonable threshold was chosen to estimate and classify the images according to their similarities to standard face .The experi-ment proved that this method is effective in eliminating mis-aligned face image effectively and is beneficial for in-creasing the reliability of a face recognition system .