激光杂志
激光雜誌
격광잡지
LASER JOURNAL
2014年
9期
54-57
,共4页
Adaboost算法%支持向量机%蚁群算法%人脸表情识别
Adaboost算法%支持嚮量機%蟻群算法%人臉錶情識彆
Adaboost산법%지지향량궤%의군산법%인검표정식별
Adaboost algorithm%SVM%ant colony algorithm%facial expression recognition
针对AdaBoost算法随着学习难度的增加导致分类器的分类效率下降、稳定性变差等问题，支持向量机在小样本中有特有优势；本文结合两种算法优势，基于蚁群算法对SVM的参数进行优化，改进了Ada-boost_SVM级联分类算法，首先提取haar-like矩形特征通过Adaboost分类器快速排出非人脸区域；用Gabor小波变换提取人脸表情特征，再结合Adaboost_SVM级联分类器进行人脸表情识别。通过对JAFFE表情库进行试验，表情平均识别率达到94.2%，检测速度有了很大提高。
針對AdaBoost算法隨著學習難度的增加導緻分類器的分類效率下降、穩定性變差等問題，支持嚮量機在小樣本中有特有優勢；本文結閤兩種算法優勢，基于蟻群算法對SVM的參數進行優化，改進瞭Ada-boost_SVM級聯分類算法，首先提取haar-like矩形特徵通過Adaboost分類器快速排齣非人臉區域；用Gabor小波變換提取人臉錶情特徵，再結閤Adaboost_SVM級聯分類器進行人臉錶情識彆。通過對JAFFE錶情庫進行試驗，錶情平均識彆率達到94.2%，檢測速度有瞭很大提高。
침대AdaBoost산법수착학습난도적증가도치분류기적분류효솔하강、은정성변차등문제，지지향량궤재소양본중유특유우세；본문결합량충산법우세，기우의군산법대SVM적삼수진행우화，개진료Ada-boost_SVM급련분류산법，수선제취haar-like구형특정통과Adaboost분류기쾌속배출비인검구역；용Gabor소파변환제취인검표정특정，재결합Adaboost_SVM급련분류기진행인검표정식별。통과대JAFFE표정고진행시험，표정평균식별솔체도94.2%，검측속도유료흔대제고。
Aiming at the fault of reduced classifier efficiency、poor stability and other issues of AdaBoost algo-rithm that caused by the increase of the learning difficulty and the unique advantages that Support vector machine has in small samples, based on ant colony algorithm to optimize the parameters of SVM, this paper improves Ada-boost_SVM cascade classification algorithm combining the advantages of Adaboost algorithm and SVM. Haar-like rectangle features are used to remove non-face by using Adaboost classifier. Gabor wavelet transformation is adopted to extract features of facial expression, and then, combined with Adaboost_SVM cascade classification to recognize facial expression. Experimental result shows that the recognition rate reaches 94.2%and the detect speed has been greatly improved through JAFFE database.
