融合小波变换与KPCA的分块人脸特征抽取与识别算法

鉴于小波多尺度变换对高维图像特征具有良好的压缩和表达能力,提出了一种融合小波变换与KPCA(核主成分分析)方法的分块人脸特征抽取与识别算法。该算法首先对人脸图像进行分块小波变换,再根据图像块的位置分布选取不同的频率分量;然后对此分量进行KPCA特征抽取,并通过对抽取到的特征进行融合来得到最终人脸鉴别特征;最后利用支持向量机分类器进行特征分类与识别。通过对ORL和Yale标准人脸图像库的实验仿真结果表明,该算法不仅在识别性能和分类速度上明显高于传统的PCA方法及融合小波特征的KPCA方法,而且对于人脸光照、姿态和表情变化均具有良好的鲁棒性。

Features Extraction and Recognition of Intersected Human Face Based on Wavelet Transform and KPCA

According to powerful compression and expression ability for high-dimensional image of wavelet multi-scale transformation,a feature extraction and recognition technology of intersected human face based on wavelet transform and KPCA(kernel principal component analysis) is proposed in this paper.With this method,the image of human faces are firstly divided into small different pieces which then being transformed with wavelet transformation algorithm.Secondly according to the positions of intersected small images coefficients of different frequency are chosen as extracted warelet features.Thirdly with KPCA the principal components of these features and then by combining these the ultiwate discriminate features are obtained.Finally the features are classified with the classifier of SVM(support vector machine).The experimental results on ORL and Yale face database show that the proposed method is superior to traditional PCA methods and KPCA methods with wavelet transformation,and it is also fairly robust to the variety of different illumination condition,face pose and expression.