融合全局和局部特征的图像特征提取方法

Research on Image Feature Exaction Method by Combining Global and Local Features

针对图像特征提取无法同时利用样本的全局和局部特征的问题,提出融合全局和局部特征的特征提取方法.该方法充分利用线性判别分析和保局投影算法分别在特征提取中保持样本全局特征和局部特征方面的优势,进一步提高图像特征提取效率.首先,引入全局散度矩阵和局部散度矩阵分别表征样本的全局特征和局部特征.然后,基于同类样本尽可能紧密,异类样本尽可能远离的思想,构造最优化问题.比较实验表明:与传统的主成分分析、线性判别分析、保局投影算法相比,文中方法的工作效率有一定提高.

With the development of application,the main problem of image feature extraction is almost no study taking both global and local features into consideration.In view of this,feature exaction approach by combining global and local characteristics（FEM-GLC）is proposed in this paper.The advantages of linear discriminant analysis（LDA）in extracting the global feature and locally preserving projections（LPP）in preserving the local feature are taken into consideration in FEM-GLC which tries to improve the efficiencies of feature extraction.In FEM-GLC,the global divergence matrix and the local divergence matrix are introduced which respectively represents the global feature and local feature.The optimization problem of FEM-GLC is constructed based on the close relation between samples of the same class and far away between different classes.The comparative experiments with PCA,LDA and LPP on the ORL dataset and Yale dataset verify the effectiveness of FEM-GLC.