摘要
为克服图像识别中传统方法需要进行高维矩阵奇异值分解的困难,提出了先局部降维再总体降维的组合变换方法.ORL人脸图像数据库的实验表明,这一方法不仅减少了运算量,而且能较好地解决人脸这样一类复杂的图像识别问题;总体来说优于传统的基于KL变换的识别方法.
To overcome the difficulty of SVD of high dimension image matrix and improve the performance of image recognition, a new method of combining the local dimension compression with the integral dimension compression is presented. Recognition experiments are implemented on the ORL face database, and the results show that the computing cost is reduced as compared with those of methods based on the KL transform, and that the method presented also has better performance in recognizing the complicated images such as human-face.
出处
《海军工程大学学报》
CAS
北大核心
2006年第2期23-26,63,共5页
Journal of Naval University of Engineering
基金
国防预研基金资助项目
关键词
特征抽取
最佳鉴别特征
KL变换
Fisher最佳鉴别方向
feature extraction
optimal discrimination feature
KL transform
Fisher optimal discrimlnatlon orientation