结合优化MGFR与二维线性降维的特征提取算法

受限于人脸姿态、光照变化等因素,通过引入多通道Gaborface表征结合基于子空间的二维双向线性降维算法,提出了一种结合优化多通道Gaborface与二维线性降维的特征提取算法。首先,采用多通道Gaborface表征(MGFR)模型对样本集进行预处理,提取不同通道下的人脸Gabor特征表示并优化选取通道融合方式而组合成新特征;再引入样本间类别信息获得改进线性二维双向特征降维算法,从而对获得的人脸表示进行特征降维与提取;最终通过最近邻分类器得到分类结果。试验结果表明,通过在AR、ORL和YALE人脸库进行对比分析,改进算法对人脸姿态等变化具有较强的鲁棒性,且较其他算法表现出了较优的识别性能。

Feature Extraction Algorithm Combining the Optimization of MGFR and Two-Dimensional Linear Dimensionality Reduction

Limited by factors such as face pose and illumination change,combined with a two-dimensional bidirectional linear dimensionality reduction algorithm based on subspace and multi-channel gaborface representation,a feature extraction algorithm combining the optimization of multi-channel gaborface representation and two-dimensional linear dimensionality reduction is proposed.Firstly,the multi-channel gabor face representation(MGFR)model is used to preprocess the sample set,and the facial Gabor feature representations in different channels are extracted and the better channel fusion methods are selected to form a new feature representation.Due to bringing in the class information between sample,the improved linear two-dimensional bidirectional feature reduction algorithm is obtained,so that the obtained face representation is subjected to feature dimension reduction and extraction.Finally,the classification result is obtained by the nearest neighbor classifier.The experimental results show that the improved algorithm is robust to changes in face pose and other performances by comparing on the AR,ORL and YALE face databases,and it has better recognition performance than other algorithms.