基于2维保局投影的人脸识别

Face Recognition Based on Two-dimensional Locality Preserving Projections

特征提取是人脸识别的一个重要研究领域,能否有效地提取判别特征是决定人脸识别算法好坏的关键。一般的人脸识别算法都是基于图像向量的,需要将2维人脸图像压缩成1维向量,这不仅破坏了像素之间原有的空间结构关系,而且转换后的向量维数过高。为了避免这种情况,提出了一种直接基于图像矩阵的人脸识别算法——2维保局投影算法。由于该算法是在保局投影的基础上进行扩展,使其可以直接面向2维图像矩阵进行处理,同时在构建相似矩阵的时候引入了样本类别信息,因而可有效地提取人脸图片的2维判别特征。另外还采用最小近邻分类器估算识别率。在AT&T人脸库的实验结果表明,与Eigenface、Fisherface以及Laplacianface算法相比,该方法具有较好的识别率。

Feature extraction is an important step of face recognition. To extract discriminant feature effectively is the key point for a good face recognition algorithm. Normally the face recognition algorithm is based on the image vector which is converted from the image matrix. A new face image feature extraction and recognition method based on two-dimensional locality preserving projections（2DLPP） was proposed in this paper. 2DLPP works directly with images in their native state- two dimensional matrices, and extracts the two-dimensional discriminant feature of face for recognition based on both the face manifolalocal structure information and the labels＇ information. The proposed method was tested and evaluated with the AT＆T face database, where the nearest neighborhood（NN） algorithm was used to construct classifiers, and the experimental results show that 2DLPP is more powerful than the PCA, LDA and LPP for face feature extraction and recognition.