摘要
针对局部三元模式提取到的人脸特征通常具有较高的维数,导致特征的紧致度不高,提出一种新的局部人脸特征提取方法——LTP子模式,并结合线性鉴别分析获得最佳的人脸局部纹理紧致特征的分类投影轴。本文在ORL和AR两个标准人脸库上测试,LTP-SP提取到的人脸特征维数不到原LTP特征的30%,但是识别性能却优于原始算法,因此算法具有较好的应用前景。
Face recognition algorithms based on Local Ternary Pattern（LTP） have the problem that the face features extracted by LTP are high-dimensionality.The LTP features are always low compacted.Aiming to resolve this problem,in this paper,we propose a new local feature extraction method called Local Ternary Pattern Subpattern（LTP-SP） approach.Analysis is employed to reduce the feature vector dimensionality,and the optimal classification projection axes of face local texture features is obtained using Linear Discriminant Analysis.Testing on ORL and AR face database,the experimental results show that the dimensionality of the proposed LTP-SP feature is only about 30% of the original LTP feature,but the recognition of the LTP-SP method is superior to the LTP method.So the proposed method has good application prospects.
出处
《吉林工程技术师范学院学报》
2012年第5期77-80,共4页
Journal of Jilin Engineering Normal University
关键词
人脸识别
局部三元模式
局部特征
维数约减
face recognition
local ternary pattern
local features
dimensionality reduction
