摘要
人脸特征的选择对识别结果起关键作用。传统上只提取较大奇异值特征作为识别特征的人脸识别方法,识别率不高,对表情和姿态变化敏感。SVD-TRIM算法选择的奇异值识别特征融合了人脸整体和局部细节特征,并采用基于"一对一"的LSSVM多类分类器分类识别。实验结果表明SVD-TRIM算法选择的识别特征对提高识别率具有较大贡献,且对光照、姿态和表情具有鲁棒性。
Feature selection of face image is the key to face recognition.The conventional method to extract algebraic features of face image based on the Singular Value Decomposition(SVD) leads to low recognition accuracy and high sensitivity to the varieties of facial expression,illumination and posture.In this paper,a novel method of features selection based on SVD-TRIM algorithm is proposed.The new features syncretize whole and part features of face image.Experimental results,based on LSSVM,suggest that the new features improve recognition accuracy and insensitivity to the facial expression,illumination and posture.
出处
《工程图学学报》
CSCD
北大核心
2010年第5期74-80,共7页
Journal of Engineering Graphics
基金
863国家高技术发展计划资助项目(2006AAJ119
2006AAJ210)