摘要
预先对人脸图像进行离散小波变换来尽量消除对识别无关的信息,以达到在提高识别率的同时有效降低时间复杂度,同时为了抑制光照的影响,还引入了一种对图像灰度进行指数衰减的图像预处理策略。通过在Yale人脸库上进行试验表明,结合以上预处理的核Fisher鉴别分析(Kernel Fisher Discriminant Analysis,KFDA)方法有着比传统KFDA方法以及当前较好的零空间KFDA方法更好的识别率。
Discrete wavelet transformation was firstly used to eliminate the information which was not related to the identification in order to improve the recognition rate and effectively reduce the time complexity. To inhibit the effects of light, a strategy to pre-process the attenaution image was introduced. The experimental results, based on Yale database face, show that the combination of the above methods to deal with the KFDA has a better performance than traditional KFDA and the current zero KFDA.
出处
《广东工业大学学报》
CAS
2009年第4期62-64,87,共4页
Journal of Guangdong University of Technology
