基于HAAR小波的关联加权LDA在人脸识别中的研究

线性判别分析(LDA)是人脸识别系统中用来降维的主要技术之一,可以运用于整个人脸图像,但却受到了小样本(small sample size,SSS)问题的限制。通过引入权值的概念,关联加权LDA(RW-LDA)方法有效地改善了小样本问题,但是,它的分类效果却不是很好。为了解决这个问题,提出了基于HAAR小波的关联加权LDA(related weighting LDA based onHAAR wavelet,HWRW-LDA)方法,在HAAR小波子带基础上,应用关联加权LDA方法,既解决了小样本问题,又改善了分类的效果。在ORL及FERET两大人脸数据库的实验结果表明,与最先进的几种方法相比较,HWRW-LDA方法具有更好的识别性能。

Research of Related Weighting LDA Based on HAAR Wavelet in Face Recognition

As is known to all of us, Linear Discriminative Analysis (LDA) is one of the principal techniques used in face recognition systems. It can be worked on the whole face image, but LDA is limited by Small Sample Size (SSS) problem. Related Weighting LDA (RW-LDA) addresses the SSS problem by using conception of weighting, however, the classification of RW-LDA is not well. To address this problem, Related Weighting LDA based on HAAR Wavelet (HWRW-LDA) which applying the RW-LDA on HAAR wavelet subband is proposed in this paper. It improved classification performance as well as addressing the SSS problem. Experiments on ORL and FERET face database clearly demonstrate this and the graphical comparison of the algorithms clearly show that proposed method gets the better recognition performance comparing with the latest approaches.