摘要
构建了一种基于核函数的典型相关分析的特征融合算法。首先,利用核函数将图像矩阵映射到核空间,再抽取同一模式的两组特征向量,在两组特征向量之间建立描述它们的相关性的判据准则函数;然后依此准则函数抽取两组典型投影矢量集;最后通过给定的特征融合策略抽取组合的典型相关特征以用于分类识别。该算法将两组特征向量之间的相关性特征作为有效鉴别信息,既可以很好地融合信息,又可以有效地去除特征之间的信息冗余,并且避免了对映射后的数据矩阵进行分解,从而简化了数据运算。在AR、PIE、ORL、Yale人脸数据库及UCI手写体数字库上的实验结果证明了该方法的有效性和稳定性。
A feature fusing algorithm based on kernel canonical correlation analysis KCCA）was constructed in this paper.First,the image data are mapped through kernel function,and then two groups of feature vectors with the same pattern are extracted and the correlation criterion function between the two groups of feature vectors are established.Secondly,two groups of canonical projective vectors are extracted according to this function.Thirdly,feature fusion for classification is done by using proposed strategy.The advantage of the proposed algorithm lies in the following aspects.Firstly,it suits for information fusion.Secondly,it eliminates the redundant information within the features,and it simplifies the computation without decomposing the mapped matrix and gains more discriminated features.The results of experiments on AR face database,PIE face database,ORL face database,Yale face database and UCI handwritten digit database show that this algorithm is efficient and robust.
出处
《计算机科学》
CSCD
北大核心
2016年第1期35-39,共5页
Computer Science
基金
国家自然科学基金项目(61170121
61202312)资助
关键词
核函数
核典型相关分析
特征融合
组合特征抽取
人脸识别
Kernel function
Kernel canonical correlation analysis
Feature fusion
Combined feature extraction
Face recognition
