[1]张震,李龙.基于加权小波分解和Fisherfaces的人脸识别算法研究[J].郑州大学学报(工学版),2014,35(03):47-50.[doi:10.3969/j.issn.1671 -6833.2014.03.012]
　ZHANG Zhen,LI Long.Face Recognition Algorithm Research Based on WeightedWavelet Decomposition and Fisherfaces[J].Journal of Zhengzhou University (Engineering Science),2014,35(03):47-50.[doi:10.3969/j.issn.1671 -6833.2014.03.012]
点击复制

基于加权小波分解和Fisherfaces的人脸识别算法研究()
分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
35
期数:
2014年03期
页码:
47-50
栏目:
出版日期:
2014-06-30

文章信息/Info

Title:
Face Recognition Algorithm Research Based on WeightedWavelet Decomposition and Fisherfaces
作者:
张震; 李龙
Author(s):
ZHANG Zhen; LI Long
School of Eleetrical Engineering, Zhengzhou University, Zhengzhou 450001 , China
Keywords:
face reeognition;  whitening;  preproeessing;  weighted wavelet decomposition;  Fisheraces;  nea-rest neighbor classifer
分类号:
TP391.41
DOI:
10.3969/j.issn.1671 -6833.2014.03.012
文献标志码:
A
Abstract:
A face reeognition algorithm that eombines weighted wavelet deeomposilion with Fisherlaces is pres.ented. Firstly, the faee image preproeessing is condueted ater whitening pretreatment, in order to get rid olthe interferenee and noise, and to balance the energy speetrum of the image. Then wavelet decomposition isused to get the weighted combination of the low frequeney eomponent and the horizontal, verlieal high frequeney component. In combination with Fisheraees, applying the linear diseriminant analysis ( LDA ) into thespaee after the PCA transformation, we solved the problems that seattering matrix within the class is singularand PCA is not conducive to sample classifieation in dimensionality reduction process. Finally, the nearestneighbor classifier is used for the classification and reeognition. Through the experiment based on ORl andYALE face database, the wavelet basis and decomposilion level is determined to be db2 wavelet and 2 levels.and the best parameters of normalized size and feature dimension are chosen to make the recognition ratereached 98. 7596 and 10096 . The comparative experiment results show that the algorithm has better recogni.tion effeet with the feature dimension of 20 ~ 70.
更新日期/Last Update: