基于块双向Fisher线性判别分析人脸识别

A block-based bi-directional Fisher linear discriminant analysis on face recognition

为解决二维Fisher线性判别(2DFLD)分析需要较多系数用以表示图像的特征阵、只考虑了图像的列间相关性从而忽略行间相关性以及作为全局特征提取方法可能会失去一些重要的局部特征等问题,提出一种基于块双向二维Fisher线性判别分析(B2DFLD)算法。首先利用块图像获取保持重要局部信息;然后基于行列双向投影,获取提取特征信息;最后计算特征阵的Frobenius距离,并进行分类。在ORL、YALE与FERET人脸数据库上进行了实验,并同传统的8种人脸识别方法比较。实验结果表明,在确定图像块大小、改变训练样本数以及特征维数的前提下,本文方法的最好识别率都高于93.08,平均误识率高于0.15,明显优于其他方法,表明本文方法对有光照、表情以及遮挡的人脸图像识别具有较高的鲁棒性。

Two-dimensional Fisher linear discriminant analysis is an important feature extraction method for face recognition.However,the method needs many coefficients to represent feature matrices of images.Moreover,it only considers the correlation of the image between columns,which ignores the correlation between the lines.To solve the above problems,this paper proposes a block-based bi-directional Fisher linear discriminant analysis algorithm.First,the block image is used to obtain the important local information.Then,the extracted feature information is obtained based on bi-directional projection.Finally,the Frobenius distance of the feature matrix is calculated and classified.We have carried out experiments on ORL,YALE and FERET face databases,and make comparison with other methods of face recognition.Under the premise of determining the size of the image block,changing the number of training samples and the feature dimensions,the best recognition rate of the proposed method is higher than 93.08,and the average error rate is higher than 0.15,which are obviously superior to those of other methods.It shows that this method has high robustness to illumination,facial expression and occlusion in face image recognition.