摘要
提出了基于分块Fisher线性鉴别(FishersLinearDiscriminant,FLD)的特征提取方法,先对图像矩阵进行分块,将分块得到的子图像矩阵直接用来构造类内和类间离散度矩阵,然后利用Fisher鉴别函数取极大值时得到的最优投影方向进行图像的特征提取。分块FLD方法是二维FLD方法的推广,该方法可以提取每一单元块的局部特征,在ORL人脸库上的实验结果表明该方法在人脸识别性能方面优于二维FLD方法。
A feature extraction technique called blocked FLD is presented. First, the original images are divided into sub-images in proposed approach. Then, between-class and within-class scatter matrixes are constructed directly using the sub-images, and when the Fisher＇s linear discriminant function reach the maximum, its optimal projection direction are derived for image feature extraction. 2D FLD is the special case of blocked FLD in which local feature can be extracted from each sub-image. To test blocked FLD and evaluate its performance, an experiment was performed on ORL human face database. The experimental results indicated that the recognition performance of blocked FLD is superior to that of 2D FLD.
出处
《科学技术与工程》
2006年第19期3107-3110,共4页
Science Technology and Engineering
