用于表情识别的半监督学习自适应提升算法  

AN ADAPTIVE BOOSTING ALGORITHM WITH SEMI-SUPERVISED LEARNING FOR FACIAL EXPRESSION RECOGNITION

作　　者：吴会丛[1] 贾克斌[2] 蒋斌[2] Wu Huicong;Jia Kebin;Jiang Bin(College of Information Science and Engineering, Hebei University of Science and Technology, Shijiazhuang 050018, Hebei, China;College of Electronic Information and Control Engineering, Beijing University of Technology, Beijing 100124 , China)

机构地区：[1]河北科技大学信息科学与工程学院,河北石家庄050018 [2]北京工业大学电子信息与控制工程学院,北京100124

出　　处：《计算机应用与软件》2016年第6期272-276,共5页Computer Applications and Software

基　　金：河北省自然科学基金项目(F2014208113)

摘　　要：针对半监督人脸表情识别算法在表情来源多样、姿态不一时准确率低的问题,在迁移学习自适应提升算法的基础上,提出一种新的半监督学习自适应提升算法。该算法通过近邻计算由训练集中的已标记样本求出未标记样本的类别,并借助Ada Boost.M1算法分别对多数据源的人脸表情样本和多姿态人脸表情样本展开识别,实现样本的多类识别任务。实验结果表明,与标号传递等半监督学习算法相比,该算法显著提高了表情识别率,且分别在多数据库和多姿态数据库上获得了73.33%和87.71%的最高识别率。To address the low recognition rate of traditional facial expression recognition algorithm with semi-supervised learning caused by diverse expressions sources and different face attitudes, we propose a novel semi-supervised learning adaptive boosting ( SSL-AdaBoost) algorithm based on transplanting learning adaptive boosting algorithm. The algorithm determines the categories of unmarked samples by calculating the marked samples concentrated in training through near neighbour, and recognises by means of AdaBosst. Ml algorithm the facial expression sample with multi-data sources and the facial expression sample of multiple attitudes respectively to realise the multi-category recognition task of samples. Experimental results show that the algorithm significantly improves the expression recognition rate in comparison with the label propagation method and many other semi-supervised learning methods. Besides, it achieves the highest recognition rate by 73. 33% on multiple databases and 87. 71% on multi-attitude database respectively.

关 键 词：人脸表情识别 半监督学习 自适应提升 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]