基于局部相位量化特征与多尺度分类的分块人脸识别

Recognition of Intersected Face Based on Local Phase Quantization and Multi-Metric Classification

本文提出一种基于局部相位量化特征与多尺度分类相结合的方法进行人脸识别,该方法首先采用LPQ算子提取分块人脸灰度图象的LPQ直方图序列(LPQHS),然后采用PCA+LDA方法对采样后的特征数据进行降维,最后根据多尺度分类的原则进行分类识别.该算法不仅能够提取人脸纹理信息,而且能够大幅度地降低训练数据量,并且数据量的维数与原始图像大小无关.在ORL标准人脸数据库上的实验表明,该方法具有较高的识别率.

This paper presents a face recognition method based on Local Phase Quantization and Multi-Metric classification.First,LPQ operator is used to extract the LPQ Histogram Sequence（LPQHS） from block grey-level face images.Second,The feature selection method based on PCA＋LDA is applied to extract feature subspace.Finally,Face recognition is realized based on the multi-metric classification principle.The proposed method can effectively extract the face texture,and can greatly reduce the amount of training data,and the dimension of the amount of data has nothing to do with the original image size.The simulation experiments illustrate that the proposed method obtains better recognition rate on ORL face database comparing to other classical methods.